2 views

Uploaded by Panda Mimi

- HSpeed Notes
- 2 9
- 1bcape2001unit2paper2.pdf
- Solution Manual for Fundamentals of Modern VLSI Devices 2n Ed - Yuan Taur, Tak Ning
- slide_12.pdf
- Entrance Loss Coefficients
- Flow Nets
- ijcerv5n4spl_01
- Channel 3
- Quiz 4
- BARRA HUECA IMPRIMIR.docx
- dipse
- 18-03-Anskey.pdf
- Rizaldi_pipeflow
- Hogy Számol a Külföldi Extrudert
- 68 Lot Phase 5 Answer Key
- Calculating the concentration of a substance
- 0625_w01_ms_2 (1)
- Examination 1 Soln (9 23 2013)
- Observe.chem.Changes8

You are on page 1of 127

Alexandros Chremos

Doctor of Philosophy

University of Edinburgh

2008

Declaration

I declare that this thesis was composed by myself and that the work contained therein

is my own, except where explicitly stated otherwise in the text.

(Alexandros Chremos)

i

To my parents

ii

Abstract

The term “soft matter” applies to a variety of physical systems, such as liquids,

colloids, polymers, foams, gels, and granular materials. The most fascinating aspect

of soft matter lies in the fact that they are not atomic or molecular in nature. They

are instead macromolecular aggregates, whose spatial extent lies in the domain 1 nm

to 1 µm.

Some of the most important examples of soft matter are polymers, which exhibit

intriguing and useful physical properties. In this work, the adsorption and self assem-

bly of linear and star polymers on smooth surfaces are studied using coarse-grained,

bead-spring molecular models and Langevin dynamics computer simulations. The

aim is to gain insight on atomic-force microscopy images of polymer films on mica

surfaces, adsorbed from dilute solution following a good solvent-to-bad solvent

quenching procedure. In the case of linear polymers, under certain experimental

conditions, a bimodal cluster distribution is observed. It is demonstrated that this

type of distribution can be reproduced in the simulations, and rationalized on the

basis of the polymer structures prior to the quench. In addition to providing insight

on experimental observations, the simulation results support a number of predicted

scaling laws such as the decay of the monomer density as a function of distance from

the surface, and the scaling of the film height with the strength of the polymer-surface

interactions.

Star polymers represent a special class of polymers, in which one end of each

linear chain is tethered to a small central core to form a single particle. The discovery

of these molecules led to the synthesis of a wide range of new materials. Their

structures are effectively considered as intermediate between those of colloids and

linear polymers. We explore the behaviour of the star polymers (which are like

“soft colloids”) in the proximity of a surface, using Langevin dynamics simulations.

A number of different measurements such as the height, radius of gyration, and

asphericity of adsorbed stars with different number of arms, are shown to provide

valuable insights on experimental findings.

The simplest soft matter systems consist of spherical, rigid colloidal particles.

iii

ABSTRACT iv

We investigated the neighbour distribution in a two-dimensional polydisperse hard-

disk fluid, corresponding physically to a colloidal monolayer. The disk diameter

distribution was defined by a power-law with the aim of realizing a scale-free near-

neighbour network. Scale-free (power-law) behaviour is found in many important

networks, for example, in transportation systems, biochemical reactions, scientific

and movie-actor collaborations, and sexual contacts. We have provided the first

example of a scale-free network in a model condensed-matter system.

Finally, we use genetic algorithms, a method for efficiently searching for minima

on energy landscapes, to investigate the ordered equilibrium structures formed

by binary mixtures of anisotropic dipolar particles confined on a plane, under the

presence of an external magnetic field. The anisotropy of the interparticle forces is

controlled by tilting the external magnetic field with respect to the plane. Initially, as

the field is tilted the structures are only slightly perturbed, but once the anisotropy

exceeds a critical value, completely new structures emerge.

Acknowledgments

I would like to thank my advisor Dr. Philip J. Camp for teaching me how to do

research. His confidence in my work was very supportive and it encouraged me

greatly. I always had the freedom to follow my own ideas, for which I am very grateful.

I would like to thank my (informal) advisor Dr. Vasileios Koutsos, who initiated

the polymer projects. The fact that I could spend time in a different department gave

me the opportunity to see things from a different perspective. Additionally, I would

like to thank Dr. Emmanouil Glynos for the endless discussions about polymers and

other subjects.

I would like to thank Prof. Christos N. Likos for giving me the chance to learn

and work with genetic algorithms during a three month visit to the Department of

Theoretical Physics, University of Düsseldorf, whose support via SFB and hospitality

are gratefully acknowledged. In addition, I am grateful to Julia Fornleitner for valuable

discussions about genetic algorithms.

Former and current colleagues from our group, namely Peter Duncan, Leonardo

Lenoci and Georg Gunzemuller, never ceased to amazed me with their actions in our

group meetings especially those abroad which were carefully planned and organised

by our advisor, Dr. Philip J. Camp.

Finally but not least I thank for the financial support of the EPSRC and the School

of Chemistry of the University of Edinburgh.

v

Contents

Declaration i

Abstract iii

Acknowledgments v

1 Introduction 1

1.1 Polymers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 Scale-free neighbour networks in polydisperse 2D fluids . . . . . . . . . 5

1.3 Colloidal monolayers with dipolar interactions . . . . . . . . . . . . . . 6

1.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 General Considerations 8

2.1 Statistical physics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.1.1 Ensembles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.1.2 Time averages and ensemble averages . . . . . . . . . . . . . . . 10

2.2 Simulation Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.2.1 Interaction potentials . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.2.2 Reduced Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.2.3 Periodic Boundary Conditions . . . . . . . . . . . . . . . . . . . . 13

2.2.4 Monte Carlo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.2.5 Molecular dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.2.6 Brownian dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.3 Polymers and Scaling Theory . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.3.1 Definition and structure . . . . . . . . . . . . . . . . . . . . . . . . 18

2.3.2 Polymer conformations . . . . . . . . . . . . . . . . . . . . . . . . 19

2.3.3 Solvent effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2.3.4 Scaling concepts of polymers . . . . . . . . . . . . . . . . . . . . . 23

2.4 Fractals and self-similarity . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.4.1 Fractal dimension and Koch curve . . . . . . . . . . . . . . . . . . 25

2.4.2 Fractal universe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

3.2 Simulation model and methods . . . . . . . . . . . . . . . . . . . . . . . 32

3.3 Results & discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3.3.1 Isolated polymer . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

3.3.2 Many polymers – good solvent . . . . . . . . . . . . . . . . . . . . 39

vi

CONTENTS vii

3.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

4.2 Simulation Model & Methods . . . . . . . . . . . . . . . . . . . . . . . . . 52

4.2.1 Simulation Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

4.3 Results & Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

4.3.1 Isolated star – Athermal solvent . . . . . . . . . . . . . . . . . . . 53

4.3.2 Isolated star - Solvent Effects . . . . . . . . . . . . . . . . . . . . . 62

4.3.3 Many adsorbed stars - Athermal solvent . . . . . . . . . . . . . . 65

4.3.4 Many adsorbed stars - Bad Solvent . . . . . . . . . . . . . . . . . . 69

4.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

5.2 Model and methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

5.2.1 Monte Carlo simulations . . . . . . . . . . . . . . . . . . . . . . . 76

5.2.2 Scaled particle theory . . . . . . . . . . . . . . . . . . . . . . . . . 78

5.2.3 Degree distribution from maximum-entropy theory . . . . . . . 79

5.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

5.3.1 Equation of state . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

5.3.2 Radical tessellation and degree distribution . . . . . . . . . . . . 81

5.3.3 Assortativity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

5.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

6.2 Simulation Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

6.2.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

6.2.2 Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

6.2.3 Genetic Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

6.3 Results & Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

6.3.1 C ≥ 1/2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

6.3.2 C < 1/2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

6.3.3 Density effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

6.3.4 Diameter ratio effects . . . . . . . . . . . . . . . . . . . . . . . . . 106

6.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

7 Conclusions 109

Bibliography 119

Chapter 1

Introduction

In the last decades soft matter has emerged as an important subfield of condensed

matter physics. The term describes a rich variety of different substances whose

common feature is that their properties are much more sensitive to mechanical

stresses and thermal fluctuations than hard materials such as aluminium or sodium

chloride. The predominant physical behaviour of soft materials occur at an energy

scale comparable with thermal energy at room temperature, where quantum effects

are unimportant. Nature’s main materials come from soft matter; most biological

materials such as blood, muscles, proteins/DNA, milk, and leaves are classifiable as

soft matter. Soft materials can also be important in a wide range of technological

applications. Since antiquity humans have used animal and plant fibres for their

basic needs such as clothing and writing (papyrus, pergamon, and paper). Today one

can find soft materials as structural and packaging materials, foams and adhesives,

detergents and cosmetics, paints, food additives, lubricants, and fuel additives.

Perhaps the most fascinating aspect of soft matter is the interesting behaviour

that arises in ways that cannot be predicted, or is difficult to predict, directly from its

atomic or molecular constituents. This is often because soft matter self-organises

into mesoscopic physical structures that are much larger than the microscopic scale

(the arrangement of atoms and molecules), and yet are much smaller than the macro-

scopic dimensions of the material. In particular some of the microscopic factors

can include a large number of internal degrees of freedom, weak interactions be-

tween structural elements, and a delicate balance between entropic and energetic

contributions to the free energy. These properties lead to large thermal fluctuations,

a wide variety of structural forms, sensitivity of equilibrium structures to external

conditions, macroscopic softness, and metastable states. The overall properties and

interactions of these mesoscopic structures determine the macroscopic behaviour of

the material. For example, the Indians of the Amazon basin collect the sap from a

1

1.1 Polymers 2

Figure 1.1: The figure shows screenshots of two types of polymers from computer

simulations. (a) Linear polymer composed by 200 monomers; (b) a star polymer

composed by 64 arms and 100 monomers per arm.

hevea tree and place it on their foot. After letting it dry for a short period of time, the

material becomes a boot. From microscopic perspective the starting point is a set of

independent, flexible polymer chains. The oxygen from the air builds in a few bridges

between the chains, which transforms the material from the liquid state to a network

structure which can resist tension. This material is known as rubber. Hence, soft

matter is an exciting field in which even a mild chemical action can induce drastic

changes on the properties of the material.

This thesis is devoted to the study of the self-organisation of some soft-matter

systems. Two types of systems are studied here: polymers and colloidal monolayers.

1.1 Polymers

Polymers represent one of the outgrowths fields in soft matter. They are polyatomic

molecules composed by small building blocks (a group of atoms), called monomers.

The most fundamental topology of polymeric structure is a linear chain as seen in

Fig. 1.1(a). One can generate new topologies by tethering the one end of chains to a d

dimensional surface, for example d = 0 (point) creates a star polymer Fig. 1.1(b) and

d = 2 (surface) creates a brush. To tether a chain at a two-dimensional surface is done

with a chemical bond, whose monomer-surface interaction strength is U À kT . One

such bond is strong enough to keep the rest of the chain tethered on the surface and

significantly reducing its mobility. This is a type of adsorption called chemisorption.

1.1 Polymers 3

Figure 1.2: (a)-(c) AFM images of linear polybutadiene (molecular mass 78.8kg/mol)

showing the patterns created after good-to-bad solvent transition occurs on ad-

sorbed of a thin film polymers on a mica surface. The surfave covergage for each

is Γ = 4.05 × 10−3 , 7.75 × 10−2 , and 4.01 × 10−1 mg/m2 respectively. (d)-(f ) Cluster-

height distributions correspoding to images (a)-(c), respectively. Pictures courtesy of

Emmanouil Glynos.

Nevertheless, one can have an adsorbed polymer on a surface without tethering it.

This can be achieved when the monomer-surface interaction strength is U ∼ kT ,

meaning that these monomer-surface contacts can easily break. Adsorption occurs

due to many monomer-surface interactions. The resulting conformational relaxation

of the polymer is driven by the competition between the entropic due to loss of

conformational freedom, and the drop in energy from binding monomers to the

surface. This type of adsorption is called physisorption. Polymer physisorption from

dilute solution has been frequently studied in the past, but to the best knowledge

of the author no previous studies have focused on the pattern formation of the

physisorbed polymers from a good solvent-to-bad solvent quenching transition. In a

collaboration with E. Glynos and V. Koutsos of the School of Engineering of University

of Edinburgh, patterns have been observed with atomic-force microscopic (AFM), in

which two types of polymers were studied, linear [1] and star polymers [2]. The AFM

images, in case of linear polymers, reveal a bimodal cluster distribution under certain

experimental conditions, see Fig. 1.2. Like in the Amazonian boots example above it

is easy to imagine what is happening. Before the good-to-bad solvent transition, the

chain spread on the surface to maximise their monomer contacts with the surface.

Additionally being in good solvent conditions the monomer-solvent interaction is

1.1 Polymers 4

Figure 1.3: High-contrast AFM images in air, 1 × 1µm2 , showing the patterns created

after good-to-bad solvent occured at adsorbed star polymers on a mica surface. From

image (a) to (c) the adsorbed amount is 0.104, 0.304 and 1.75 mg/m2 , respectively.

Pictures courtesy of Emmanouil Glynos

attractive, which makes the chain soluble. However, once entered into bad-solvent

conditions the monomer-solvent interactions becomes repulsive. Hence, the chain

collapses into a structure that minimises the surface contact with the solvent. For

low surface density each chain behaves as if it was an isolated chain, since it does

not feel the presence of the other chains. Therefore once the good-to-bad solvent

transition occurs each chain will collapse into itself leading to a cluster distribution

dominated by single collapsed chains. Nevertheless, for higher surface density, an

extra entropic term affecting the chain conformation is needed, which describes the

competition for space between different chains. When the the good-to-bad solvent

transition occurs at high surface density the chains are already close to each other

by having monomer-monomer contacts or even entanglements for long chains, and

that leads to the creation of many-chain clusters. One should expect that there is an

intermediate regime in which both clusters and single collapsed chains are observed.

In chapter 3 we demonstrate that this type of distribution can be reproduced with

computer simulations and we rationalise the observed AFM structures on the basis

of the polymer structures prior to the quench.

AFM experiments have also been performed with star polymers, see Fig. 1.3. Star

polymers are a special case of branched polymers where a star is composed by f

flexible polymer chains tethered to a core. f is called functionality. The functionality

governs the star’s behaviour from a linear chain for f = 1 and 2 to a colloidal particle

for f À 1. In other words, a star polymer is a hybrid between polymer-like objects

and colloidal particles, hence bridging two different domains of physics. A great

interest has been evoked recently on how stars behave in the proximity of a substrate.

For example it is interesting to understand how stars can act as depleting agents

of colloidal particles, in other words to prevent colloidal aggregation and with an

1.2 Scale-free neighbour networks in polydisperse 2D fluids 5

Figure 1.4: Three examples of the kinds of scale-free networks. (a) A vizualization

of the network structure of the Internet at the level of autonomous systems. (b) A

map of interacting proteins. (c) A social network, in the case of sexual contacts. The

images were taken from the Refs. [6, 7].

ultimate goal to assemble the colloids into ordered structures [3, 4, 5]. However, it

is still an unexplored region on how star polymers behave in the proximity of an

attractive surface. In chapter 4 we discuss how a surface can affect the conformation

of a physisorbed star. Measurements from computer simulations of the fraction of

monomers bound to the surface, height, radius of gyration, and asphericity provide

valuable insights on a star polymer’s conformations. We rationalise on the basis of

the results and it will allow us to map a star’s behaviour (from linear to soft-colloid)

based on the number of arms and their length, and the properties of the surface. The

mapping the behavior of an adsorbed star polymer on a surface provides us valuable

insights and can act as a guide to predicting a star’s conformation on a surface for

future experiments.

fluids

systems, for example, as models of colloidal monolayers, froths and foams, and

of packing and segregation in granular materials. In particular we focus on the

neighbour network of a two-dimensional polydisperse hard-disk fluid with a power

1.3 Colloidal monolayers with dipolar interactions 6

law diameter distribution. The aims are twofold. First we want to understand how a

diameter distribution with a slowly decaying tail, as it happens to power laws and

in contrast with exponential distributions, will affect the neighbour network. It is

tempting to speculate that if the diameter distribution follows a power law, then

perhaps the network distribution will also vary in a similar way. The assumption is

that for any given disk the number of immediate neighbours will be proportional to

its circumference and hence its diameter. And second if the neighbour network does

vary with a power law then does it also have the properties of scale-free networks?

Scale-free networks are characterised in qualitative terms by a small number of highly

connected vertices called “hubs” and a large number of low connected vertices, see

Refs. [6, 7]. Some examples of real scale-free networks are illustrated in Fig. 1.4. The

degree distribution (i.e. the probability of finding a vertex with k connections) varies

with a power law, a necessary but not sufficient condition for a network to be scale-

free. Many physical systems have been identified as scale free, for example the World

Wide Web, the Internet, metabolic and protein-protein interactions, sexual contacts,

collaborations among scientists. But what makes scale-free networks fascinating is

that they have self-similar behaviour. For example, in a self-similar network where the

vertices are connected with edges (connections of links) there is no possible way to

assess whether a vertex has an above or below average number of connections.1 This

happens because power law distributions with slow decaying tails have infinite mean,

thus there is no characteristic scale to use for comparison. In chapter 5 we examine

the polydisperse hard-disk fluid by using constant-pressure Monte Carlo simulations.

Additional measurements are discussed in the chapter such as the assortativity of

the network, which is found to be positive, meaning that vertices of equal degree are

connected more often than in a random network. Finally, the equation of state is

determined and compared with the prediction from a scaled-particle theory.

In the final part of the thesis, we look at two-dimensional binary mixtures of dipolar

colloids. A popular description of colloids is that they are particles with the simplest

and highest possible symmetry (sphere), and without any additional structure on the

mesoscopic length scale, see Fig. 1.5. It is possible experimentally to insert within

the colloid a magnetised substance and thus effectively create a colloid carrying a

point magnetic dipole. The importance of this is for example creating new type of

polymers where its monomers are colloids; or one can use external fields and take

1

A finite size network will be bounded by a minimum and maximum degree, which can give a finite

mean. The self-similarity manifests between the bounds.

1.4 Summary 7

2.7µm, at the interface between water and a mixture of decane and undecane in the

through. The picure is taken from the Ref. [8].

create well disperse colloidal particles in the substance of interest. In chapter 6 we

are interested in what are the global minimum-energy configurations of a binary

mixture of dipolar colloids confined to a flat interface, and with a magnetic field is

applied perpendicular direction polarising the particles and inducing long range

repulsive interactions between them. We use genetic algorithms to efficiently search

the global minimum of energy landscapes; genetic algorithms are described in the

methodology of chapter 6.

1.4 Summary

This thesis is organised as follows. In chapter 2 we briefly summarise key concepts in

statistical mechanics, computer simulations, polymers and fractals that underpin

this thesis. In chapters 3 and 4 we study the self-assembly of physisorbed linear and

star polymers, respectively. In chapter 5 we look at the behaviour of the neighbour

network of a two-dimensional polydisperse hard-disk fluid with slow decaying diam-

eter distribution. In chapter 6 we examine ordered equilibrium structures formed

by binary mixture of anisotropic dipolar particles confined on a plane, under the

presence of an external magnetic field. Finally, in chapter 7 we conclude this thesis.

Chapter 2

General Considerations

there were two ways one could understand Nature. The first was to perform experi-

ments, for example by recording the time for a falling apple from different heights to

hit the ground. The second method is by developing a theory. Theory takes math-

ematics and constructs a framework in which a model can be built and studied.

Thus, in the case of a falling apple (or more precisely the orbits of planets) Newton

developed the theory of gravity, which explains how the apple falls, and planets orbit

around the sun rather than crushing each other. Theories can be tested by comparing

their predictions with experimental results. Disagreement between experimental

results and theoretical predictions can lead to a new theory, as happened in the 19th

century when Mercury’s peculiar trajectory was first observed and led to Einstein’s

theory of general relativity.

The development of computers has provided an alternative mode of research

in science. Computer simulations can be used as a linking bridge between theory

and experiments. Mathematical in nature, computer simulations provide a new

metaphor for the laws of Nature, the algorithms. The first systematic explorations

with computers occurred in 1952, with Metropolis being interested in having the

Los Alamos MANIAC machine tried against a wide spectrum of mathematical and

physical problems. However, simulations, or to be more accurate algorithms, which

are the heart of computer simulations, are not new discoveries. The ancients used

algorithms to gain insights into difficult mathematical or physical problems; one

example is the Euclidean algorithm, which determines the greatest common divisor

in any Euclidean domain (i.e. integers). Simulations can easily examine different

theories and visualise the results thus providing valuable insights. Furthermore,

they can aid in interpreting experimental results, and in devising new experimental

strategies.

8

2.1 Statistical physics 9

systems frequently use statistical mechanics. Statistical mechanics is the discipline

that connects the microscopic variables (positions, momenta, and interactions) of

a physical system to its macroscopic ones (temperature, pressure, etc.). Many sim-

ple problems in statistical mechanics are exactly solvable, but only a handful of

non-trivial problems can actually be solved exactly. Other problems can be tack-

led by using successive approximations, and perturbative expansions; computers

play an important role in such work, for instance in calculating virial coefficients,

understanding polymer conformations, and so on.

Polymer physics is founded upon the principles of statistical mechanics, which

studies the structural and dynamical properties of polymers. It is a relatively old

field (starting in 1920), but since then it has become a fundamental part of statisti-

cal mechanics providing great insights into understanding polymers like DNA and

designing new exotic materials. One of these insights was that polymers are finite

fractals, which corresponds to geometrical objects whose geometry is invariant on

all scales.

In this chapter we briefly summarise the key concepts in statistical mechanics,

computer simulations, polymers and fractals that underpin this thesis.

Statistical mechanics is a probability theory which provides mathematical tools

for dealing with systems of many particles. It is able to predict statistics (averages,

fluctuations) for the macroscopic properties (such as for instance, pressure and

temperature) of a many-body system from its microscopic properties, such as inter-

actions.

2.1.1 Ensembles

Statistical mechanics groups many-body systems into ensembles. An ensemble in

statistical mechanics consists of a very large (theoretically infinite) number of copies

of a system, all characterised by the macroscopic varibles. There are four ensembles

that are commonly used. (a) The micro-canonical ensemble is a group of systems

characterized by a fixed number of particles N , total energy E , and in volume V .

This ensemble is often denoted as the NV E ensemble and consists of copies of an

isolated system. (b) The canonical ensemble consists of many-body systems with

equal number of particles N , which are at thermal equilibrium with a heat bath at

temperature T , and at the same volume V . This ensemble exchanges energy with the

2.1 Statistical physics 10

heat bath and is denoted as the NV T ensemble. (c) The grand-canonical ensemble

is a collection of identical systems at equilibrium with an external reservoir with

which they exchange both energy and particles. This ensemble is refered to as the

µV T ensemble, where µ is the chemical potential controlling the average in the

number of particles. (d) In the isothermal-isobaric ensemble, all the systems have

the same number of particles N and maintain the same temperature T and pressure

P . It is refered to as the N P T ensemble and it plays an important role in chemistry

as many chemical systems are prepared and studied under conditions of constant

temperature and pressure.

represented in phase space. In the case of a three-dimensional system of N particles,

the values assumed by the variables rN and pN at any time define a point in the

6N -dimensional phase space. A set of points in phase space linked by dynamical

equations of motion define a trajectory.

If A rN , pN is a function of the 6N coordinates and momenta, its time average

¡ ¢

1

Z τ£

A rN , pN dt

¡ ¢¤

〈A 〉t = lim (2.1)

τ→∞ τ 0

A different averaging procedure based on the concept of ensemble can also be intro-

duced. The average properties of a system can be obtained from a single “snapshot”,

rather than tracking the system’s configurations through time. Hence, we use a func-

tion describing the distribution in phase space at time t of the phase points relative to

a certain statistical-mechanical ensemble. This function is called probability density,

f rN , pN , t . Under equilibrium conditions f is independent of time and ensemble

¡ ¢

ZZ

A rN , pN f rN , pN , t d rN d pN

¡ ¢ ¡ ¢

〈A 〉e = (2.2)

If the average in eq. 2.1 is performed over a long time, and the system eventually

flows through all its possible microstates, then the ensemble averages become the

same as the time averages, that is

〈A 〉t = 〈A 〉e (2.3)

This equivalence is known as the ergodic theorem and it is believed to hold for all

2.2 Simulation Techniques 11

In this section, we discuss a few computational tricks that are of great practical

importance for an efficient simulation program. Additionally, we discuss some fun-

damental concepts in computer simulations such as how Monte Carlo and molecular

dynamics algorithms work.

We describe our model of the system of interest by a function which gives the po-

tential energy for a given configuration of the particles in the system. Usually, the

assumption is made that the potential is pairwise additive. That is, we assume that

the total potential is just the sum of potential for each pair of particles, which is

expressed as a function of their relative positions (and possible orientations). This

is justified because three-body (or higher order) terms in the potetial give a small

contribution into the description of the overall system. More details can be found in

Refs. [9, 10].

Perhaps the simplest model of a particle is to regard it as an impenetrable hard

sphere and it is represented by the potential

(

∞ r <σ

U (r ) = (2.4)

0 r >σ

where r is the distance between the particles and σ is the particle diameter. The

hard core interaction which prevents the particles from overlapping provides a good

representation of the Pauli repulsion which operates at short distances and has its

origin in the exclusion principle. Another, widely used pair potential is the Lennard-

Jones potential,

σ 12 ³ σ ´6 ¸

·³ ´

U (r ) = 4ε − (2.5)

r r

where r is the distance between the particles, σ is the distance parameter and ε is

a parameter giving the depth of the minimum in the potential. At large distances,

this has the asymptotic 1/r 6 dependence of the van der Waals interaction, and the

1/r 12 term approximately describes the hard-core repulsion when the particles come

close together. This potential provides an excellent description for argon and it is

frequently used for many other molecules.

2.2 Simulation Techniques 12

Figure 2.1: The diagrams display the the hard-sphere potential (top) and Lennard-

Jones potential (bottom).

beyond which the interactions are negligible. This means that any two particles

further apart than this distance do not need to be considered when calculating the

total potential energy, speeding up the simulation. The discontinuity in the potential

can cause problems, however. A discontinuous potential means an infinite force

(since F = −∇U ), which can lead to numerical instabilities. One way to avoid this is to

shift the potential in the vicinity of the cut-off by a function which brings it smoothly

to zero.

units in terms of the potential parameters of the system under study. For example,

for Lennard-Jones fluid, with the pair potential given in Eq. 2.5, distances can be

expressed in terms of the parameter σ, and energies can be expressed in terms of ε.

We also use the mass of the particles in the system as a fundamental unit (i.e. m = 1).

From these fundamental definitions, we can obtain other quantities in reduced units.

Some reduced units for the Lennard-Jones potential are listed in Table 2.1.

The most important reason for using reduced units is that there are many different

states in real units that corresponds to the same state in reduced units. The use of

reduced units means we do not have to repeat the simulation for each of these states.

Another advantage is that values in reduced units will be of order unity, hence we

2.2 Simulation Techniques 13

Distance r ∗ = r /σ

Density ρ ∗ = ρσ3

Energy E ∗ = E /ε

Temperature T ∗ = k B T /ε

Pressure P ∗ = P σ3 /ε

Time t ∗ = (ε/mσ2 )1/2 t

avoid the potential problems of using the very small numbers of atomic systems on a

computer. The use of reduced units also simplifies the potential, allowing us to set

the parameters of the potential to one. We do not need to choose specific values of

these parameters.

In a simulation of finite number of particles, a significant fraction of particles will be

at the edges of the system. The environment of these particles will be very different to

the environment of a particle within the bulk system. In order to reduce this problem,

periodic boundary conditions are normally employed. We make copies of the system

in all directions, see Fig. 2.2. When a particle leaves the central simulation cell at

one side, its periodic image enters from the other side, thus conserving the number

of particles in the central simulation cell. Technically, every particle interacts with

every particle in all cells, but often the minimum image convention is used. When

calculating the force and the interactions on a pair of particles, the only interaction

that is considered is that due to the nearest of the periodic images of the particular

particle.

Monte Carlo methods aim to generate a trajectory in phase space which samples

from a chosen statistical-mechanical ensemble. The ensemble average of a certain

quantity A is then calculated via

1X

〈A 〉 = Ai exp(−βE i ) (2.6)

Q i

exp −βE i is the partition function, β = 1/k B T and exp −βE i is the

P ¡ ¢ ¡ ¢

where Q = i

Boltzmann factor which weights every accessible state i of energy E i . However, the

2.2 Simulation Techniques 14

number of accessible states in a large system can be huge thus making the exact

evaluation of the average in eq. 2.6 unfeasible with current computers. Moreover,

certain states (those with small Boltzmann factor) contribute very little to the average,

and hence should be omitted in an efficient calculation. To overcome these problems,

importance-sampling is employed. In this scheme the only configurations to be

sampled are those that make a significant contribution to the partition function

and ensemble averages. This task is achieved by generating a Markov chain of

configurations in which each state is generated by “perturbing” the preceding one in

the chain, and accepted as a new configuration only if the detailed balance condition

is satisfied.

A sufficient (but not necessary) detailed-balance condition is that for old (o) and

new n states

J (o → n) = J (n → o) (2.7)

where J (o → n) is the probability flux from a state o to a new state n during a given

simulation step. The flux J (o → n) can be factored into three terms

where p (o) is the probability of being in state o (which at equilibrium should follow

the Boltzmann distibution), α (o → n) is the probability of generating the trial move

from o to n, and acc (o → n) is the probability of accepting the trial move from o and

n. By combining eqs. 2.7 and 2.8 the detailed-balance condition can be rewritten as

= · (2.9)

acc (n → o) acc (o → n) p(o)

2.2 Simulation Techniques 15

p(n)

where p(o)

is obtained based on the statistical ensemble we work on. In the case of

NV T ensemble the ratio can be expressed as the Boltzmann factor of the energy

difference:

−1 −βE (n)

p(n) Z NV T e

= −1 −βE (o) = e −β[E (n)−E (o)] (2.10)

p(o) Z NV Te

where E (n) and E (o) are the system energies for the new and old state, respectively.

Z NV T is the partion function defined as the sum of all the energy states, E i , and

read as Z NV T = e −βE i . Now if we assume that α (o → n) = α (n → o), then eq. 2.9

P

becomes ³ ´

acc (o → n) = min 1, e −β[E (n)−E (o)] . (2.11)

in the canonical (NV T ) ensemble will proceed as follows. The system is prepared

in some initial configuration. A particle is chosen randomly and displaced within

the simulation cell. If the trial move results in a decrease of the system energy, then

the move is accepted and the particle positions updated. If the trial move results

¡ ¢

in an increase of energy, instead, the move is accepted with probability exp −β∆E ,

where ∆E is the difference in system energy associated with the trial move. This

is implemented in a computer simulation by generating a random number in the

interval (0, 1) and accepting the move only if the random number is less than the

¡ ¢

value exp −β∆E . These steps are repeated for the subsequent randomly chosen

particles. After a large number of trial moves, simple unweighted averages over

the visited states can be calculated since each state is sampled with a probability

proportional to its Boltzmann factor.

Molecular dynamics (MD) simulations is a classical technique to compute equilib-

rium structural and dynamical properties of a many-body system.

In this technique each particle moves according to Newton’s laws of motion,

which are integrated numerically. The positions and velocities of the particles are

updated every time-step and the equations of motion integrated to generate a trajec-

tory in phase space. This procedure is repeated for a number of time-steps required

to equilibrate the system and calculate accurate time averages.

A typical MD simulation will proceed as follows: positions ri and velocities vi of

the particles at time t are stored in the computer. The force acting on each particle

is calculated from the potentials (Fi = −∇i U ) and then the equation of motion,

Fi = m i ai , is integrated over a finite time-step ∆t according, for instance, to the

Verlet algorithm to get the position of the particles in the next time-step, t + δt [9]

2.2 Simulation Techniques 16

ri (t + δt ) = ri (t ) − ri (t − δt ) + ai (t )δt 2 (2.12)

ri (t + δt ) − ri (t − δt )

vi (t ) = (2.13)

2δt

The numerical integration of the equations of motion using the Verlet algorithm and

its variants generates errors of order δt 4 in the positions and δt 2 in the velocities.

more accurately the velocities (the errors is of order δt 4 ) and it reads as

1

ri (t + δt ) = ri (t ) + vi (t )δt + ai (t )δt 2 (2.14)

2

1

vi (t + δt ) = vi (t ) + [ai (t ) + ai (t + δt )] δt . (2.15)

2

The original Verlet algorithm may be recovered by eliminating the velocities. The

algorithm only requires storage of ri , vi and ai and its implementation involves two

stages, with the force evaluation in between. First, the new positions at time t + δt

are calculated using eq. 2.14 and then the forces and accelerations at time t + δt

are computed. At this point the velocities at time t + δt are available. The origin of

eq. 2.15 comes by calculating the velocities at mid-step i.e. at time t + 21 δt

1 1

vi (t + δt ) = vi (t ) + ai (t )δt (2.16)

2 2

and at time t + δt

1 1

vi (t + δt ) = vi (t + δt ) + ai (t + δt )δt . (2.17)

2 2

Combining eqs. 2.16 and 2.17, we obtain eq. 2.15.

Overall, the Verlet algorithms for most MD simulations are perfectly adequate.

The success of Verlet algorithms relies on their simplectic character, meaning that it

has a time-reversal symmetry and long-term energy conservation.

ties. A typical MD program simulates a microcanonical ensemble (constant NV E ).

MD simulations can also be performed in constant-temperature ensembles, such

as NV T (canonical ensemble), and N P T (where P is the system pressure). In these

cases the temperature T is kept constant by coupling the system to a heat bath. A

detailed review of constant-temperature MD methods is given in Refs. [9] and [11].

2.3 Simulation Techniques 17

Brownian dynamics (BD).

the solvent and the slower motion of the solute, such as in the case of polymers,

amphiphile and colloidal suspensions, the simulation time can became prohibitively

long. In fact, the short time steps needed to capture the fast motion of the solvent,

and the long runs needed to study the structural evolution of the solute, can make

simulations very time consuming, especially in those cases in which the fast solvent

modes are not of interest. BD is a mesoscale simulation technique in which solvent

particles are replaced by a combination of stochastic and frictional forces.

At the very core of this technique there is the integration of the Langevin equation

in order to generate trajectories of the solute particles in which we are interested,

letting friction and random terms to mimic the solvent interactions with the solute.

The Langevin equation in this situation is written

Fi = −∇i U − ξm i vi + Wi (2.18)

where vi is the velocity of particle i , U is the total potential energy, ξ and Wi are

the friction coefficient and the random force representing the effects of the solvent.

The random force is, in a typical BD simulation, Gaussian noise, which obeys the

fluctuation-dissipation theorem

〈Wi (t )〉 = 0

〈Wi (t ) · W j (t 0 )〉 = 6mk B T ξδi j δ(t − t 0 ). (2.19)

The Langevin equation 2.18 can be integrated numerically, for example using

a Verlet-like algorithm. The friction coefficient ξ is related to the single-particle

diffusion coefficient D accordingly to Einstein’s equation

kB T

ξ= (2.20)

mD

where k B is the Boltzmann’s constant and m is the mass of the diffusing particle.

the flow of the solvent around them, and the interaction between two particles

due to solvent structure, can be included in this simulation scheme. Details are in

Refs. [9, 12, 13].

2.3 Polymers and Scaling Theory 18

The development of human civilisation has been dictated by the availability of mate-

rials. In fact, the discovery of a new material can have such drastic effects as to lead

to the decline of an old civilizations and give the rise to a new one. For this reason

history is divided into eras named after the primary materials used; the Stone Age,

the Bronze Age, and the Iron Age. Similarly, we can say that in the 20th century we

entered the Polymer Age.

In this section we will discuss the definition of a polymer, and review some

universal properties of linear chains.

Polymers are macromolecules (polyatomic structures) consisting of many elementary

units, called monomers, and they are connected to each other by covenant bonds.

The entire structure of a polymer is generated through the process of polymerisation,

the process by which elementary units are covalently bonded together. The number

of monomers in a polymer molecule is called the degree of polymerisation, N . For

example, polyethylene’s repeating unit is −CH2 −CH2−, since the monomers used in

their synthesis are ethylene, CH2 =CH2 , see Fig. 2.3.

The definition of a polymers is generic and includes a huge variety of polymers

with distinct structural and architectural features, such as DNA and polypetides;

these features change the physical properties of the molecules. For instance both the

boiling point and the melting point increase rapidly with the number of backbone

monomers. We briefly describe some of the ways of categorising these structures. The

first way of categorising a polymer is by the number of different types of monomers

in the molecule. If a polymer contains only one type of monomer, then it is called a

homopolymer, otherwise it is called a heteropolymer. Heteropolymers have a whole

range of subcategories depending on the number of different types of monomers

and their position in the molecule. For example, for two types of monomers, say

A and B, then you can have different sequences of monomers, e.g. alternating ...-

A-B-A-B-A-B-A-..., random ...-A-A-B-B-B-A-B-..., diblock ...-A-A-A-B-B-B-... chains.

2.3 Polymers and Scaling Theory 19

Figure 2.4: Examples of polymer architectures (a) linear; (b) ring; (c) star (d) H; (e)

comb; (f) ladder (g) dentrimer; (h) randomly branched.

The second was by which one can categorise polymers is by their size. Oligomers

are polymers consisting of a small number of monomers (generally, less than 20);

polymers may contain between 20 and 10 billion (for the longest known chromo-

some) monomers. Third, another important way of characterising of polymeric

systems is by their architecture. Types of polymer architectures include linear, ring,

star-branched, H-branched, comb, ladder, dendrimer or randomly branched, see

Fig. 2.4. The properties of branching and the formation of networks (via a high degree

of crosslinking) make polymers useful as soft solids (e.g. erasers, tyres).

look first to the possible conformations. The simplest model that describes a linear

polymer is a freely joined chain. The model represents the polymer as a chain of n + 1

identical monomers with a constant bond length |ri | = l , see Fig. 2.5. Moreover, we

make an assumption that the chain is in an ideal state, meaning that there are no

interactions between the monomers separated by a large number of bonds along the

chain, so that |i − j | À 1.

We define the end-to-end vector, which is the sum of all n bond vectors ri in the

chain:

n

X

Rn = ri (2.21)

i =1

Different configurations of the chain will have different bond vectors and hence

different end-to-end vectors. However, the ensemble average of Rn over all possible

states by either considering many chains or many different configurations of the

2.3 Polymers and Scaling Theory 20

〈Rn 〉 = 0 (2.22)

which signifies that there is no preferred direction in this ensemble. The simplest

non-zero average is the mean-square end-to-end distance:

*Ã ! Ã !+

n n n X

n

〈R 2 〉 =

X X X

ri · rj = 〈ri · r j 〉 (2.23)

i j i =1 j =1

If all bonds have the same length l = |ri |, then the scalar product can be represent

in terms of the angle θi j between bond vectors ri and r j

ri · r j = l 2 cos θi j (2.24)

n X

n

〈R 2 〉 = l 2 cos θi j

X ®

(2.25)

i =1 j =1

In the freely joined chain model there are no correlations between the directions

of different bond vectors, thus cos θi j = 0 for i 6= j . There are only n non-zero terms

®

quite simple:

〈R 2 〉 = nl 2 (2.26)

However, in a typical polymer chain, there are correlations between bond vectors

(especially between neighbouring ones), so 〈cos θi j 〉 6= 0. So one can group these

2.3 Polymers and Scaling Theory 21

ideal flexible chain composed by effective bond of Kuhn length, b.

1X n X n

cos θi j

®

Cn = (2.27)

n i =1 j =1

where the coeffiecient C n , called Flory’s characteristic ratio, is the average value of

the nj=1 cos θi j over all main-chain bonds of the polymer. Now eq. 2.25 can be

P ®

reduced to

n X

n

〈R 2 〉 = l 2 cos θi j = C n nl 2

X ®

(2.28)

i =1 j =1

All real polymers have C n > 1, meaning that the local correlations are significant

i.e. cos θi j 6= 0, which means that the local alignment of the bonds favours a

®

parallel orientation. The physical origins of these local correlations between the

bond vectors are restricted bond angles and steric hindrance. All models of ideal-

chain polymers ignore steric hindrance between monomers separated by many

bonds. This has as a result for a chain with large numbers of main chain bonds

(n → ∞) their characteristic ratios to saturate C ∞ = limn→∞ C n . Thus, 〈R 2 〉 can be

approximated for long chains by

〈R 2 〉 ∼

= C ∞ nl 2 (2.29)

The values of the characteristic ratios of some common polymers are listed in Ta-

ble 2.2.

The Flory characteristic ratio allows a simple description of all ideal polymers by

an equivalent freely joined chain. The equivalent chain has the same mean-square

end-to-end distance 〈R 2 〉 and the same maximum end-to-end distance R max as the

actual polymer, but has N freely-joined effective bonds of length b, see Fig. 2.6. This

2.3 Polymers and Scaling Theory 22

Table 2.2: Characteristic ratios, and Kuhn lengths for common polymers

1,4-Polybutadiene (PB) −(CH2 CH=CHCH2 )− 5.3 9.6

Poly(ethylene oxide) (PEO) −(CH2 CH2 O)− 6.7 11

Atactic polystyrene (PS) −(CH2 CHC6 H5 )− 9.5 18

effective bond length b is called the Kuhn length. Therfore, the Kuhn length for an

equivalent freely joined chain is given by

〈R 2 〉 C ∞ nl 2

b= = (2.30)

R max R max

where R max is the contour length of the equivalent freely joined chain.

Eq. 2.30 renormalizes the monomers, and holds for all flexible linear polymers in

the ideal state (N À 1), with all chemical-specific characteristics contained within

the Kuhn monomer (i.e. Kuhn length). This makes the properties of flexible polymers

independent of the local chemical structure, and thus universal.

can significantly alter configuration of the polymer. There are three types of solvents,

good-solvent, ideal (or θ-solvent) and bad-solvent. In good solvent conditions the

interactions between the solvent molecules and the polymer monomers are attrac-

tive. This means that the polymer is soluble and the effective monomer-monomer

interaction is repulsive. The chain conformation can be described as a self-avoiding

walk, which is a random walk on a lattice that never visits the same site more than

once. In an ideal solvent the net interaction between solvent molecules and polymer’s

monomers is zero. The polymer conformation resembles a random walk.1 In the case

of the bad-solvent conditions we have an attractive effective monomer-monomer

interaction and the polymer collapses into a globular configuration for minimising

the surface contact between the polymer and solvent.

1

The monomers of a real chain in ideal-solvent do not overlap, but the statistics of the chain are

the same with that of a random walk.

2.3 Polymers and Scaling Theory 23

A simple and excellent scheme was devised by Flory for obtaining valuable insights

on the conformations of linear chains. We briefly describe his method here, for more

details see Refs. [14, 15].

are uniformly distributed within volume R d with no correlations between them.

Note that the argument holds for an arbitrary dimensionality d . The probability of

a second monomer being within the excluded volume v (in good solvent v > 0) of

a given monomer is the product of the excluded volume v and the number density

of the monomers in the pervaded volume of the chain N /R d . The energetic cost of

being excluded from this volume is kT per exclusion of kT v N /R d per monomer. For

all N monomers in the chain, this energy is N times larger

N2

F int ≈ kT v (2.31)

Rd

The Flory estimate of the entropic contribution to the free energy of a real chain is

based on the fact that the entropic elasticity obeys the Hooke’s law, and thus it is

equal to the the energy required to stretch an ideal chain to end-to-end distance R:

R2

F ent ≈ kT v (2.32)

N b2

The total free energy of a real chain in the Flory approximation is the sum of the

energetic interaction and the entropic contributions:

N2R2

µ ¶

F = F int + F ent ≈ kT v d + (2.33)

R N b2

The minimum free energy of the chain (obtained by setting ∂F /∂R = 0) gives the

optimum size of the real chain in the Flory theory, R = R F :

Ã !

∂F N2 RF

= kT −d v d +1 + 2 =0

∂R R N b2

F

d 2 3

R Fd +2 ≈ vb N (2.34)

2

The Flory theory leads to a universal power law dependence of polymer size R F

2.3 Polymers and Scaling Theory 24

Figure 2.7: Stretching a linear chain by applying a force f at both ends of the chain.

The characteristic length ξ represents the region in which within it the chain statists

are unperturbed by the stretching.

3

R F ∼ N ν with ν = (2.35)

d +2

interesting to note that ν is independented of v, thus being a universal exponent.

One can work out with similar arguments the values of ν in different solvents, for

more details are in Ref. [15]. The other values of ν for d = 3 are 1/2 corresponding to

an ideal chain (or θ-solvent conditions where the excluded volume v = 0) and 1/3

describing the collapsed state of the chain when it enters into bad-solvent conditions.

The predictions of the Flory theory are in good agreement with both experiments

and other more sophisticated theories. However, its success lies on cancellation of

errors, because both the entropic and energetic terms in eq. 2.33 are overestimated.

Nevertheless, Flory theory is useful because it is simple and provides a reasonable

answer.

b size, under tension. The stretching is done by applying a force of magnitude f at

both ends of the chain. In this problem the only characteristic lengths involved are

(a) the Flory radius R F , and (b) the length ξ = kT / f , which defines a tension blob of

size ξ containing g monomers each, such that on length scales smaller than ξ the

chain statistics are unperturbed, while on larger length scales the chain is a string of

independent blobs, see Fig. 2.7. The number of monomers per blob, g , is related to ξ

by the Flory law ξ ≈ bg ν .

2.4 Fractals and self-similarity 25

The end-to-end distance R f is the product of the tension blob size ξ and the

number of these blobs N /g

N f b 1/ν−1

µ ¶

Rf ≈ ξ ≈ Nb . (2.36)

g kT

This shows that a chain in good solvent (ν = 3/5) has an elastic response which is

nonlinear, 1/ν − 1 = 2/3, than in the case of an ideal chain (ν = 1/2). In both types of

solvents a stretched chain will have fewer possible configurations, but in good solvent

conditions it has fewer possible configurations to lose. For this entropic reason it

returns a smaller stretching force, but increases more rapidly with extension.

This concept of scaling, that is to divide a chain of Kuhn monomers into blobs

that each contribute of order kT to the free energy, further simplifies the problem at

hand.

As demonstrated in section 2.3.1 polymers can adopt many different conformations

that due to their size (the largest DNA molecules has 1010 bonds) can span through

a wide range of length scales. But at the same time, they can be coarse-grained

to high degree without losing the overall structural information. There is another

feature in polymers that is related with the above characteristics, and that is most

polymers are self-similar, or finite random fractals. In this section we will explore

self-similar objects and some basic fractal properties by looking at polymers, and

other examples.

Any polymer follows the Flory law, eq. 2.35 R ∼ N ν , where ν takes the values 3/5, 1/2,

and 1/3 in three dimensions for good, ideal and bad-solvent conditions, respectively.

The values of ν characterise the conformations the chain takes, for example in a bad

solvent (ν = 1/3) the chain collapses into a globular structure. To explain this we

briefly look to a mathematical example.

Consider a regular fractal, called the Koch-curve, see Fig. 2.8. To generate it

one has to start with a section of a straight line and then divide it into three equal

subsections. Then on the top of the middle subsection an equilateral triangle is

drawn with its bottom side erased. Thus we end with four segments of equal length,

with total length 4/3. Iterating the process infinitely many times results in a curve of

infinite length, which is everywhere continuous but nowhere differentiable. Another

2.4 Fractals and self-similarity 26

Figure 2.8: On the left an illustration shows the Koch curve after different number

of iterations and on the right a display of the self-similar feature to a high order

approximation to the Koch curve. (Both images were taken from Google search with

related keywords)

striking feature is that if we magnify any section of the curve we will observe the same

structure, see Fig. 2.8. This is an example of a fractal.

To understand this further we consider the following example. We have an infinite

long straight rod of thickness (or diameter) H and it has uniform density. Now we

draw a sphere of radius r around the rod, as seen in Fig. 2.9(a) and ask the question:

how the amount of mass, m, within the sphere changes as function of the radius of the

sphere. When the sphere is much bigger than the thickness of the wire (r À H ) then

the rod is effectively an one-dimensional object and so m ∼ r . If however the sphere

is smaller than the thickness (or diameter) of the rod (r < H ) then the mass within

the sphere will scale with the volume, thus m ∼ r 3 . The dependence of the mass m

of the part of the wire inside the sphere of radius r is sketched in Fig. 2.9(a). Thus

we can say that the rod is one-dimensional on length scales much larger than its

thickness r À H and three-dimensional on smaller scales r < H . Another important

point to make is the relation between m and r in the form of m ∼ r D , where D is a

critical exponent (also called Hausdorff dimension).

Now we apply similar arguments for the case of the Koch curve. We plot two circles

of diameter 2r equal to the lengths of the lengths of two consecutive generations,

see Fig. 2.9(b). The radius of the circles changes by a factor of 3 (r 1 = 3r 2 ), while the

number of segments within the circles changes by a factor of 4 (n 1 = 4n 2 ). Similar

with the previous example we assume that the number of segments within a cicle is

n ∼rD (2.37)

2.4 Fractals and self-similarity 27

Figure 2.9: (a) The diagram shows how the amount of mass m of the part of the wire

inside the sphere of radius r as function of r . Both axes have logarithmic scales. (b)

Comparing the number of segments part of the Koch curve for two different circle

radius, r 1 and r 2 .

n 1 = cr 1D = c(3r 2 )D (2.38)

n1 = 4n 2 = 4cr 2D (2.39)

where c is a proportionality constant for eq. 2.37. Equating eqs. 2.38 and 2.39 will

give

(3r 2 )D = 4r 2D (2.40)

which can be solved for the Hausdorff dimension of the Koch curve.

ln 4

D= ≈ 1.26... (2.41)

ln 3

The value of D represents the dimensionality of the object; here it is close to one,

which is somehow comforting because an infinity long curve is, in some sense, more

than a one-dimensional object without being a two dimensional area.

Coming back to polymers the Flory value ν is related with the Hausdorff dimen-

sion in the following way

1

ν≡ (2.42)

D

The Flory value of ν for a two-dimensional polymer chain in good solvent conditions,

is ν = 3/4, and its Hausdorff dimension is D = 1.333...; this is quite close to the

Hausdorff dimension for the Koch curve. However, polymers are finite random

fractals, quite different from the Koch curve which is an example of a regular fractal.

We emphasise two key differences. (a) A polymer is a self-similar object because

if its smaller sections are magnified, they look similar to the whole chain. Unlike

2.4 Fractals and self-similarity 28

regular fractals the magnified sections do not look exactly like the whole chain, but

only on average. (b) Polymers can span many length scales but remain finite. Their

self-similar features manifest themselves between the length scales of the monomer

and that of the whole chain. This comes in contrast with regular fractals which are

self-similar on all length scales.

Mathematical fractals remain invariant to any change of scale. Polymers are not the

only objects that manifest this property. That is because the underlying concept

of fractals is symmetry – by symmetry we mean an invariance against change –

and symmetry is everywhere in Nature. For example a widely noticed symmetry

is the mirror symmetry, that is, invariance against flipping sides. Nature’s most

fundamental laws of physics, such as Newton’s law of gravitation, have an exact

mirror symmetry: there is no difference between left and right. Nevertheless the non-

conservation of parity in radioactive decay – that is the violation of point symmetry

in the weak interactions needs to be noted. Emmy Noether was the first to point out

the connection between the symmetries in the fundamental laws of physics and the

conservation of energy and momentum. In chapter 6 we observe that patterns of

two-dimensional binary mixtures of dipolar colloids remain unchanged over wide

ranges of density. The cause of this invariance in this example lies in the interactions

between the colloids, which scales with a power law ∼ 1/r 3 . As we have seen in

previous examples, power laws are related with fractals, due to their slow decaying

tail spanning many (length) scales. In chapter 5 we discuss scale-free networks which

relate and explain how physical systems that might look very different, such as

protein networks, sexual contact networks and the Internet, can be governed by the

same laws.

Chapter 3

polymers on surfaces

3.1 Introduction

Polymers near to, or adsorbed on, surfaces exhibit useful and interesting proper-

ties. Adsorbed polymers find application in colloid stabilisation, [16, 17] nanoscale

surface patterning, [18] friction modification, [19, 20] DNA microarrays [21] and

adhesion. [22] Polymers can be attached to appropriate surfaces either through

chemisorption/grafting (i.e., anchoring by chemical bonds) or by physisorption (i.e.,

chain attachment by van der Waals interactions). For a weakly adsorbing surface

the physisorption and resulting conformational relaxation of the chain is driven by

the competition between the entropic repulsion due to the loss of conformational

freedom and the drop in energy from binding monomers to the substrate. Earlier

investigations have focused on thin-film polymer blends, [23, 24, 25] block copoly-

mer micelles adsorbed on surfaces, [26, 27, 28] end-grafted polymers chemisorbed

on surfaces, [29, 30, 31] and several other complex systems. The simple case of ho-

mopolymer chains physisorbed on a substrate has been studied with simulations of

free [32, 33, 34, 35, 36, 37, 38] and tethered chains, [39, 40, 41] and through theoretical

approaches; [14, 42, 43] experimental studies, however, are scarce.

In recent work by our groups, we studied the physisorption and self-assembly

of star [2] and linear [1] polymers on smooth surfaces using using atomic-force

microscopy (AFM). In a typical experiment, a polymer solution was prepared in good

solvent at concentrations below the critical overlap volume fraction (φ∗ ), resulting

in well-separated chains in solution and hence precluding any strong degree of

structural ordering within the polymer component. Polymer (sub-)monolayers were

formed by exposing a smooth surface (such as highly ordered pyrolytic graphite

29

3.1 Introduction 30

(a) (b)

Figure 3.1: (Colour online) (a) AFM image of linear polybutadiene (molecular

mass 78.8 kDa) adsorbed on mica from toluene at a surface concentration of

7.75 × 10−2 mg m−2 . The image size is 6 × 6 µm2 . (b) Cluster-height distribution

corresponding to the AFM image in (a).

or mica) to the polymer solution. The surface was then placed in a good solvent

bath for several hours and extensively rinsed with good solvent to remove any non-

adsorbed polymer. Finally, the samples were dried gently under a stream of nitrogen

and subsequently imaged in air by AFM in tapping mode to investigate the resulting

structures from this good solvent-to-bad solvent ‘quench’. Depending on the polymer

molecular weight, architecture, and concentration, very different surface structures

can be obtained. Chapter 4 is focused on the case of star polymers, where it was found

that the functionality (number of arms) and concentration of star polymers controls

a crossover between ‘polymer’ and ‘soft-colloid’ regimes, being distinguished by

characteristic cluster topologies, sizes, and surface coverages [2]. Using the same

experimental procedure, we have now studied the adsorption of linear polybutadiene

(PB) on to mica from toluene. A report of this investigation is in preparation [1], but

for the purposes of the current work, we present one key experimental observation

on which we will aim to gain insight using computer simulations. In Fig. 3.1(a)

we present an AFM image of a freshly cleaved mica exposed for 30 minutes to a

toluene PB solution (molecular weight 78.8 kg mol−1 ) leading to a surface coverage

of 7.75 × 10−2 mg m−2 . This image shows two distinct types of clusters. In Fig. 3.1(b)

we show the corresponding bimodal cluster-height distribution, with most probable

heights of 1 nm and 6 nm. In this chapter we aim to reproduce, and gain insight

on, the development of bimodal cluster distributions using Langevin dynamics

simulations of coarse-grained ‘bead-spring’ models of linear polymers.

In the experimental quenching procedure described above, the microscopic

3.1 Introduction 31

structures of the polymer solutions in contact with surfaces prior to quenching must

control the nature of polymer adsorption. Hence, the behaviour of polymers in

good-solvent conditions and near a surface is of considerable interest. In the past,

adsorption and depletion-layer effects have been the subjects of many experimental,

theoretical, and computer-simulation studies. Of particular relevance to the current

work is the seminal study by de Gennes focused on the monomer volume-fraction

profile φ(z) as a function of the perpendicular distance from the surface z [44]. De

Gennes considered a semidilute solution of chains in contact with a weakly attracting

wall, with the wall-monomer interaction of a range comparable to the monomer

size a. In the semidilute regime, the polymer volume fraction in solution φb > φ∗ ,

meaning that there are overlapping chains. Near the wall, chains are physisorbed

through a small number of monomers, leading to the formation of loops with charac-

teristic dimension D > a. Well away from the wall, the characteristic length scale is

the bulk correlation length ξb ∼ φ−3/4

b

[14], which in the semidilute regime is com-

parable to the polymer radius of gyration RG in bulk solution. Three regimes of z

can be identified: the proximal regime z ' a < D where φ(z) is dictated by the short-

range interactions with the wall; the central regime D < z < ξb in which “no other

length enters in the problem" [44] meaning that if ξ[φ(z)] ' z and ξ(z) ∼ φ(z)−3/4

then φ(z) ∼ z −4/3 ; and the distal regime z > ξb where [φ(z) − φb ]/φb ∼ exp (−z/ξb ).

The structure in the central regime is ‘self-similar’ or ‘scale-free’, because there is

no characteristic length scale (unlike in the proximal or distal regimes, which are

characterised by a and ξb , respectively). The existence of a self-similar structure

characterised by an exponent of − 43 has been confirmed experimentally by neutron

scattering [45] and by neutron reflectivity [46, 47, 48]. Results from Monte Carlo sim-

ulations of lattice polymer models are also consistent with this behaviour [49, 50, 51].

An incidental result of the current work is a demonstration that a coarse-grained,

off-lattice polymer model can reproduce this self-similar structure, and with the

correct exponent.

In this chapter we report a simulation study of adsorbed linear-polymer films.

We use Langevin dynamics simulations of coarse-grained ‘bead-spring’ models to

gain insight on the results from polymer adsorption experiments. The outline of the

study is as follows. Firstly, we study the properties of isolated adsorbed polymers

(vanishing surface coverage). This situation has been considered many times before

(see, e.g., Ref. [38]), but for the purposes of comparison with the case of finite surface

coverage, we reconsider specific single-molecule structural properties for the par-

ticular coarse-grained models being employed. Next, we deal with many polymers

on a surface under good-solvent conditions, corresponding to the prelude to the

bad-solvent quench in experiments. Of particular interest are simulation measure-

3.2 Simulation model and methods 32

ments of φ(z) and the comparison with de Gennes’ theoretical predictions. Finally,

we simulate the good solvent-to-bad solvent quench, and its effects on the structure

of the polymer film. Specifically, we identify under what conditions a bimodal cluster

distribution (such as those seen in experiments – Fig. 3.1) should be expected. The

chapter is organised as follows. Section 3.2 contains details of the coarse-grained

polymer model, and the simulation methods. Results for isolated polymers and many

polymers are presented in section 3.3. Section 3.4 concludes the chapter.

by ‘springs’. The physical motivation for such a model stems from the fact that

correlations between monomers die off beyond a characteristic length, called the

Kuhn length b. Hence, if a number of contiguous monomers are rendered by a single

bead of dimension b, then the scaling properties of the chain on length scales larger

than b will be left invariant [14, 52, 53]. Such bead-spring models of linear and star

polymers were first introduced and employed in simulations by Grest and co-workers

[54, 55, 56, 57]. In this work, N beads of equal mass m are connected to form a chain

using a non-linear finitely extensible (FENE) potential between neighbouring beads,

given by Ã !

1 2 r2

VFENE (r ) = − kR 0 ln 1 − 2 . (3.1)

2 R0

Here r is the bead-bead separation, R 0 is the maximum possible (bonded) bead-

bead separation, and k is a spring constant. In this study we use parameters from

earlier work [55], namely R 0 = 3σ/2 and k = 30ε/σ2 ; ε and σ are the energy and range

parameters, respectively, for the non-bonded interactions to be defined next.

The non-bonded interactions operate between all pairs of beads, and are de-

rived from a composite potential devised by Steinhauser [58]. The potential is a

combination of three terms. First we write the purely repulsive Weeks-Chandler-

Andersen (WCA) potential [59], which is a Lennard-Jones potential cut and shifted at

the position of the minimum, r min = 21/6 σ:

h¡ ¢ ¡ σ ¢6 i

σ 12

(

4ε r − r + ε r ≤ r min

VWCA (r ) = (3.2)

0 r > r min

To represent the attractive interactions, the WCA potential is shifted back down in

3.2 Simulation model and methods 33

(

−λε 0 < r ≤ r min

VSW (r ) (3.3)

0 r > r min

where λ reflects the solvent quality (to be discussed below). To interpolate the

potential smoothly between −λε at r = r min and 0 at a cut-off distance r cut > r min , we

add the term

(

1

2 λε cos (αr 2 + β) − 1 r min < r ≤ r cut

£ ¤

Vcos (r ) = (3.4)

0 r > r cut

2

+ β = π and αr cut

2

+ β = 2π. The cosine form of

the potential also means that dVcos /dr = 0 at r = r cut . Following earlier work [58], we

choose r cut = 3σ/2, for which the appropriate parameters are

4π

α = p

3

= 3.173 072 868 (3.5)

9−4 2

9

β = 2π − α = −0.856 228 645. (3.6)

4

in Fig. 3.2. The parameter λ controls the depth of the potential well at r = r min ,

and provides a convenient measure of the solvent quality. In a good solvent, the

effective bead-bead interactions are purely repulsive; this corresponds to λ = 0. In

a bad solvent, the bead-bead interactions are attractive, and this behaviour can be

modelled with λ = 1; this corresponds to an attractive well of depth ² which sets the

molecular energy. θ-solvent conditions – under which the chain statistics are very

similar to those for an ideal (non-interacting) chain – are reproduced by λ = 0.646

[58].

For simulations involving a surface, an additional effective bead-surface potential

is used, [57] based on integrating the Lennard-Jones interactions arising from a

homogeneous distribution of sites within the surface. The potential is

2πεs 2 ³ σ ´9 ³ σ ´3

· ¸

Vs (z) = − (3.7)

3 15 z z

where z is the perpendicular distance of the bead from the surface, and εs controls

the strength of the bead-surface attraction. In our simulations we define εs in terms

of basic energy parameter ε by defining the dimensionless ratio ε∗s = εs /ε. In practice,

we concentrate on the values in the range 0.4 ≤ ε∗s ≤ 1.0.

For simulating the bead-spring polymer chains, we used Langevin dynamics

3.2 Simulation model and methods 34

with λ = 0.7, showing the contributions from VWCA (r ) + VSW (r ) (blue) and Vcos (r )

(red).

In addition to the conservative forces arising from the interaction potentials de-

scribed above, each bead will feel random and frictional forces mimicking the solvent

surrounding the bead. Thus the equations of motion for bead i are given by

d2 r i dr i

m 2

= −∇r i V − mΓ + Wi (t ) (3.8)

dt dt

where Γ is the friction coefficient, Wi (t ) describes the Brownian forces of the solvent

P

acting on the bead, and V = i < j Vi j is the total interaction potential. Wi (t ) is

represented by Gaussian white noise satisfying the fluctuation-dissipation theorem

[9]

〈Wi (t ) · W j (t 0 )〉 = 6k B TmΓδi j δ(t − t 0 ) (3.9)

leads to a diffusion constant for an isolated bead of D 0 = k B T /mΓ. Further details

are given in Ref. [54] and in chapter 2.

Simulations were performed in an L × L × H cuboidal box with periodic boundary

conditions in all three directions and the minimum-image convention applied. The

box dimension in the z direction was set to a large, but finite, value of H = 200σ, and

the surface was represented by a structureless, L ×L ×l solid slab with a thickness l no

less than the maximum range of interaction between beads. To control the surface

bead density, L took on values of 80σ through to 180σ, which were always large

3.3 Results & discussion 35

enough for polymers in their natural conformations to avoid interacting with their

own periodic images. The simulation conditions mean that the polymers are at finite

density within the simulation cell, and so there is an equilibrium state where the

polymers are adsorbed. In principle, the polymers could adsorb on either face of the

slab, but they cannot interact with each other because l is larger than the interaction

range, and H is much larger than typical polymer dimensions (as measured by the

radius of gyration, RG ); hence, the two surfaces are essentially isolated from one

another. In practice, initial configurations were prepared by placing the polymers

on one face of the slab, and all subsequent measurements were made for that one

surface. Simulated properties are reported here in reduced units defined in terms

of m, ε, and σ. The equations of motion were integrated using the velocity-Verlet

p

algorithm with a time-step δt = 0.007τ, where τ = mσ2 /ε is the basic unit of time.

In all cases, the reduced temperature k B T /ε = 1, and the reduced friction coefficient

Γτ = 1.

(i) the behaviour of isolated polymers on surfaces with various solvent qualities

(with 0 ≤ λ ≤ 1); (ii) the behaviour of many polymers on surfaces in good-solvent

conditions (with λ = 0), corresponding to the experimental situation before the good

solvent-to-bad solvent quench; and (iii) many polymers in bad-solvent conditions

(with λ = 1) corresponding to the post-quench situation probed in AFM experiments.

We have studied three different chain lengths (N = 50, 100, and 200 beads), a range

of different surface-energy parameters ε∗s = εs /ε, and in the many-polymer cases, a

variety of surface coverages (to be defined in section 3.3.2). The number of beads in

each of the longest chains is of the same order of magnitude as the number of Kuhn

monomers in the smallest chains studied in experiments. Such coarse-grained, bead

spring models are known to reproduce faithfully the semi-quantitative properties

of polymers, and so despite the small lengths of our chains, we anticipate that the

various trends seen in our results will be of relevance to experiments on ultra-thin

polymer films after a sudden change in the solvent quality. Our choices for ε∗s are

based achieving a suitable degree of surface adsorption. We have not attempted

the difficult problem of determining this effective interaction parameter from first

principles; this would involve using atomistic representations of the surface, polymer

and solvent in order to determine the direct and solvent-mediated forces acting

between the polymer and surface.

3.3 Results & discussion 36

The average conformation of a polymer in dilute solution is well understood, and has

an isotropic globular shape defined by a radius of gyration RG , which scales with the

number of monomer units N like RG ∼ N ν where ν is a characteristic exponent equal

1 1 3

to 3

in a bad solvent, 2

in a θ-solvent, and 5

in a good solvent [14]. In the proximity

of a surface, the number of available polymer conformations is reduced, leading

to a decrease in entropy. Adsorption occurs when the surface-energy parameter εs

exceeds a certain critical value, εcs , signaling that the energetic contribution to the

free-energy from polymer-surface interactions becomes significant [14, 60]. The

accompanying change in polymer conformation can be interpreted as the order

parameter a type of second-order phase transition at ε∗s = εcs [14, 60]. It is useful to

define the dimensionless variable κ

which measures the critical distance from the critical point. We also define the scaling

variable y,

y = κN φ (3.11)

polymer adsorption. [44, 42] Through scaling theory, one can identify four regimes

of adsorption depending on the values of y and κ. (a) For a repulsive wall, y < 0, the

chain trivially remains away from the surface. (b) Near the critical point, y ' 0, the

chain is equally likely to be found at the surface as it is in the solution. (c) When y À 1

with N large and κ small, adsorption is nonetheless favored because the sum of the

interactions of the individual beads with the surface outweighs the entropic penalty

penalty of the chain being near the wall. This situation is called weak coupling limit.

[44] (d) For κ > 1 the monomers are strongly attracted to the surface and the chain

lies flat. This is called the strong coupling limit. [44] Cases (a)-(c) have been studied

extensively in simulations of the particular situation where each chain is tethered

to the surface by one end; [40, 41] case (d) has been studied in simulations of free

chains. [36, 37, 38] Earlier simulations using the same type of bead-surface potential

(3.7) suggest that, in reduced units, the critical surface-energy parameter εcs /ε ∼ 0.1.

[41] The values of ε∗s used in the current simulations (reported below) correspond to

the strong-coupling regime.

All isolated-polymer simulations began by placing a linear chain in good-solvent

conditions (λ = 0) close enough to the surface for adsorption to occur. Once the chain

had adsorbed, the solvent quality was adjusted by changing λ to the desired value.

3.3 Results & discussion 37

The molecule was equilibrated for around 106 δt , and then properties were measured

over a production run of 5 × 106 δt . Figures 3.3(a) and (b) shows the probability

density p(z) of finding beads at a perpendicular distance z from the surface, from

simulations of isolated polymers consisting of N = 100 beads, with various values

of ε∗s and λ. (The density profiles are reported in this form – normalized so that

R∞

0 p(z)dz = 1 – to aid comparison with later results for many polymers at finite

densities.) In all cases there is either a local minimum or a point of inflection in

p(z) at z ' 1.2σ, and the position of these features was taken as a distance-based

criterion for assessing whether a particular bead is ‘bound’ to the surface or not.

Note that from equation (3.7) the minimum bead-surface potential energy is located

p6

at z/σ = 2/5 ' 0.86; this distance corresponds to the positions of the primary peaks

in p(z).

In Fig. 3.4 we show the average bound fraction 〈Φa 〉 and the average maximum

height 〈h〉 as functions of λ for different chain lengths and surface-energy parameters

ε∗s . With N = 50, no apparently stable adsorption occurs for ε∗s < 0.6, while for the

longer chains with N = 100 and 200, adsorption occurs when ε∗s ≥ 0.4. That is

because for ε∗s = 0.4 we are close to weak coupling regime, meaning that the sum

of individual bead’s enthalpic contributions with the surface for shorter chains is

not enough keep them on the surface. The bound fraction 〈Φa 〉 is the fraction

of beads within interaction range of the surface defined using the distance-based

criterion z ≤ 1.2σ. For all ε∗s , 〈Φa 〉 remain insensitive to N . This is in agreement

with theory and simulations, since for κ > 1 the bound fraction scales Φa ∼ N 0

[61]. It also remains insensitive to solvent quality. There is though only a weak

monotonic decrease with increasing λ; this is due to the polymers bunching up to

optimize the attractive bead-bead interactions, at the cost of bead-surface contacts.

Unsurprisingly, for a given λ, increasing ε∗s leads to a greater bound fraction.

In general, the average maximum height 〈h〉 for all systems with ε∗s ≥ 0.6 shows

a very weak dependence on λ, there being only a slight hint of an increase as the

bad-solvent conditions (λ = 1) are approached; this is due to the ‘bunching up’ of the

beads, to take advantage of their mutual attractive interactions. But on the whole,

the strong bead-surface interactions keep the polymers quite flat on the surface, with

small, short-lived ‘loops’ and ‘tails’ appearing as beads lift off the surface as a result

of thermal fluctuations.

Different behavior is observed in those systems with N = 100 and 200 beads, and

ε∗s = 0.4, in which 〈h〉 clearly decreases with increasing λ. This, again, is due to the

‘bunching up’ of the polymer chain, much like an accordion being compressed. The

difference here, though, is that with such a weak bead-surface interaction parameter,

the polymers possess only a small number of contacts with the surface, leading to

3.3 Results & discussion

Figure 3.3: (Color online) The probability distribution p(z) of finding a bead at a perpendicular distance z from the surface. All results are

for chains with N = 100 beads. (a) Isolated chain with ε∗s = 0.4 and 0 ≤ λ ≤ 1. (b) Isolated chain with λ = 0 (good solvent) and 0.4 ≤ ε∗s ≤ 1.0.

(c) Isolated chain compared to many chains (with densities ρ ∗ = 0.6, 0.8, and 1.0), all with λ = 0 (good solvent) and ε∗s = 1.0.

38

3.3 Results & discussion 39

Figure 3.4: (Color online) The bound fraction 〈Φa 〉 (top) and the average maximum

height 〈h〉 (bottom) against solvent quality λ for isolated linear chains with, from

left to right, N = 50, 100, and 200 beads, and with different surface-interaction

parameters ε∗s .

the formation of large, long-lived ‘loops’ and ‘tails’ oriented perpendicular to the

surface. When λ is increased, the loops and tails contract, leading to a reduction in

the height of the polymer; but with a weak bead-surface interaction, this process

occurs without the loops flattening out and forming new contacts with the surface.

priate simulations with λ = 0 were initiated by preparing configurations with many

‘curled up’ polymers on a surface, and equilibrating for around 106 δt . For chains

of N = 50, 100, and 200 beads, we initially placed 200, 100, and 50 chains on the

surface, respectively, leading to the same total number of beads in each case. Follow-

ing equilibration, we performed a production run of 2.5 × 106 δt . The adsorption is

measured by the equilibrium surface bead density, defined in terms of the number of

beads Nads belonging to those chains with at least one bead-surface contact, defined

using the distance-based criterion z ≤ 1.2σ. Note that Nads is, in general, greater

than the number of beads actually bound to the surface. The reduced surface bead

density is ρ ∗ = Nads σ2 /L 2 . By placing a fixed number of chains on surfaces of various

sizes, we simulated surface densities, up to that corresponding to the critical overlap

concetration. In other words, approached the semidilute regime within the adsorbed

film. During equilibration runs near the critical overlap concetration, some of the

polymers were seen to desorb as the polymer film approached the steady state.

3.3 Results & discussion 40

Figure 3.5: (Color online) The bound fraction 〈Φa 〉 (top) and the average maximum

height 〈h〉 (bottom) against surface bead density ρ ∗ for many linear chains with,

from left to right, N = 50, 100, and 200 beads, in good solvent (λ = 0) and with

different surface-interaction parameters ε∗s . Wherever possible, the corresponding

isolated-polymer results are shown at ρ ∗ = 0.

We first consider the average bound fraction 〈Φa 〉 and maximum height 〈h〉,

presented in Fig. 3.5. Results are shown as functions of the surface bead density ρ ∗

for chains of N = 50, 100, and 200 beads with bead-surface interaction parameters

0.4 ≤ ε∗s ≤ 1.0. For comparison, points for isolated chains are shown at ρ ∗ = 0, the

effective density in this case. In all cases, 〈Φa 〉 decreases with increasing ρ ∗ . This

may be explained by the entropic penalty associated with a reduced number of

molecular conformations due to crowding; this effect becomes more pronounced as

the surface density is increased. With purely repulsive bead-bead interactions, there

is no additional energetic gain upon adsorption (above the bead-surface interaction)

to offset the growing entropic penalty. Hence, it is more favorable for the some beads

to lift off the surface to ease crowding. Some additional observations are that for a

given N and ρ ∗ , 〈Φa 〉 increases with increasing ε∗s ; and that for a given ρ ∗ and ε∗s , 〈Φa 〉

is essentially independent of N . We note that 〈Φa 〉 has been examined in experiments

on linear-polymer films [62], but these were conducted with chemisorbed molecules,

as opposed to the physisorbed molecules considered in this work. Chemisorption

reduces sorbate mobility, and hence reduces the opportunity for reorganization. In

addition, molecules can be irreversibly chemisorbed through fewer surface contacts

than those required for physisorption. Both of these effects lead to relatively low

bound fractions, as compared to the results reported here.

Examples of the structural differences between weak and strong surface param-

eter cases are illustrated in the simulation snapshots shown in Fig. 3.6. These are

3.3 Results & discussion 41

Figure 3.6: (Color online) Top views (top) and side views (bottom) from simulations

of chains with N = 100 beads in good solvent (λ = 0). The surface dimensions are

130σ × 130σ. (Left) ε∗s = 0.4, ρ ∗ = 0.39, Nads = 6600. (Right) ε∗s = 1.0, ρ ∗ = 0.59,

Nads = 10000. Figures were prepared using Pymol (http://pymol.sourceforge.net/).

from simulations of chains with N = 100 beads. With ε∗s = 0.4 and at ρ ∗ = 0.39, the

polymers form ‘loops’ and ‘tails’, orientated away from the surface; with ε∗s = 1.0 and

at ρ ∗ = 0.59, the polymers are flat on the surface, despite the high density.

Variations in the average maximum height 〈h〉, shown in Fig. 3.5, correlate with

those in 〈Φa 〉. As the bound fraction decreases, the film height increases, reflecting

the build-up of the polymer film. Scaling theory predicts that the height 〈h〉 scales

like εν/(ν−1)

s where ν is the characteristic exponent [14, 15, 63]. For polymers in good

solvent, ν = 35 and so 〈h〉 ∼ ε−3/2

s . This applies to isolated polymers, and to many poly-

mers when the surface concentration is much greater than the bulk concentration.

Since in all cases we have an effective bulk density of zero, the scaling law should

be observed. For each system showing significant adsorption under good-solvent

conditions, we fitted 〈h〉 to the function

h0

〈h〉 = ∗ 3/2

(3.12)

(εs )

and plotted the quantities 〈h〉/h 0 on a single graph, as shown in Fig. 3.7. The results

should collapse on to the curve (ε∗s )−3/2 ; they are indeed broadly consistent with

the predicted scaling. Note that the critical surface-energy parameter is expected

to be unimportant in this analysis, because we are working in the strong-coupling

3.3 Results & discussion 42

Figure 3.7: (Color online) Scaling plot of the maximum height 〈h〉 against the surface

interaction parameter ε∗s for polymers in good-solvent conditions (λ = 0). A reduced

density of ρ ∗ = 0 corresponds to isolated polymers. The theoretical prediction [14, 15,

63] is that 〈h〉 ∼ εν/(ν−1)

s , which for good-solvent conditions (ν = 35 ) gives 〈h〉 ∼ ε−3/2

s .

h 0 is the constant of proportionality from Eq. (3.12).

regime. From the fit shown in Fig. 3.7, it is clear that the critical value of ε∗s would be

small as compared to those used in the simulations. Indeed, attempts to fit critical

parameters led to values of no more than 0.1, in reduced units, but with relative

statistical uncertainties approaching 100%.

The effects of the surface-interaction parameter on the conformations of poly-

mers in good-solvent conditions can be characterized in terms of the radius of

gyration RG defined by * +

1 X N X N

RG2 = |r i − r j |2 . (3.13)

N 2 i =1 j >i

a surface, but the average value defined above is sufficient for the current purposes.

Figure 3.8 shows the ratio γ = RG2 (many)/RG2 (isolated), where RG2 (many) is for many

polymers made up of N = 200 beads in good-solvent conditions and at finite density,

and RG2 (isolated) is the corresponding value for an isolated polymer on a surface

(and therefore at an effective density of ρ ∗ = 0). Results are shown for systems with

various surface-interaction parameters. For a given ε∗s , increasing the density causes

a decrease in γ, reflecting a crowding effect due to neighboring polymers. For a given

ρ ∗ , increasing ε∗s causes a flattening of the polymers, and hence an overall reduction

in the average dimensions.

3.3 Results & discussion 43

Figure 3.8: (Color online) RG2 for many polymers, divided by RG2 for an isolated

polymer, against surface bead density ρ ∗ , for polymers with N = 200 beads in good-

solvent conditions and with various surface-interaction parameters ε∗s .

The probability density p(z) for the many-polymer case is shown in Fig. 3.3(c).

Specifically, we show results for chains consisting of N = 100 beads in pure solvent,

with λ = 0, a fixed surface-interaction parameter ε∗s = 1.0, and surface bead densities

ranging from ρ ∗ = 0 (isolated chain) to ρ ∗ = 1.0. The proximal regime, identified by

de Gennes, [44] is dominated by the bead-surface interactions, and in this case covers

the range 0 ≤ z/σ ≤ 2; the two peaks can be interpreted as arising from two ordered

layers on the surface. For an isolated chain, p(z) dies off very rapidly beyond z ' 2σ.

At finite densities, p(z) dies off rapidly at large distances, roughly corresponding to

distal regime. Under the same conditions, there is an intermediate, central regime

in which p(z) should vary like z −4/3 in good-solvent conditions [14]; our simulation

results appear to be consistent with this scaling law. For the system with ρ ∗ = 1.0, the

central regime covers the range 2 ≤ z/σ ≤ 10; approximate ranges of the proximal,

central and distal regimes for this system are indicated in Fig. 3.3(c). In general, the

theoretical scaling predictions should apply to long chains and to adsorption from

semidilute solutions. The experimental [1] and simulation conditions correspond

more closely to an adsorbed film with a pure sovent; in addition, the simulated chains

are relatively short. Our results approach the expected z −4/3 scaling as the density is

increased due to the conditions near the surface are beginning to resemble those in a

film formed by adsorption from semidilute solutions. Note that this is consistence

even for strong adsoprtion, because the polymers extend to distances of the same

order the Flory radius, R F ∼ N ν , with large loops [42]. We emphasize that this study

3.3 Results & discussion 44

was not focused on observing the predicted scaling, but it is comforting that our

simulation results are at least consistent with the theoretical predictions [44, 42, 43];

coarse-grained, off-lattice models of polymers in good solvent appear to form de

Gennes’ ‘self-similar carpet’ [43].

here, is the quench from good-solvent to bad-solvent conditions, corresponding

to rinsing with solvent and then drying in nitrogen / air. In our simulations, we

mimic this step by starting simulations from well equilibrated configurations with

λ = 0 (good solvent) and then instantaneously switching to λ = 1 (bad solvent). We

then re-equilibrate the system for 2 × 106 δt , during which time the system was seen

to reach an apparent steady state. In our experiments, the surface structures are

apparently stable for periods of at least ten days. In our simulations no restrictions

exist in the lateral directions, e.g. surface roughness, thus allowing the polymers

slowing to diffuse on the surface. In earlier simulations of polymers films on surfaces,

apparently metastable structures have been observed for periods of time that might

extend towards experimental timescales [64]. Simulations therefore provide valuable

insights on the experimental images. Nevertheless, it is easy to imagine that the true

equilibrium state –if accessible– would correspond to a single, large (hemisperical)

polymer droplet [32, 33, 34]; so the observed behavior in our simulations might

best described as metastable. This is because the diffusion rate of an adsorbed

polymer chain in bad solvent is not only lower than in good solvent, bad also inversely

proportional to the number of beads [35]. Thus, within the simulation timescale,

clusters may not diffuse sufficiently far in order to form a putative, single-droplet

equilibrium structure. In any case, the structures we observe are apparently static on

the timescales accessed in the simulations.

Figure 3.9 shows examples of equilibrated simulation configurations before (λ =

0) and after (λ = 1) the quench, for systems of polymers each made up from N = 200

beads, with surface-interaction parameters ε∗s = 0.4, and at various densities. In

good-solvent conditions, the polymers are in extended conformations, but in bad-

solvent conditions they collapse to form globular clusters to optimize the attractive

bead-bead interactions. At high density (ρ ∗ = 0.81) the quench induces extensive

clustering, resulting in a small number of large clusters. At intermediate density

(ρ ∗ = 0.49), a mixture of single chains and large clusters is in evidence, with the

single chains in the minority. At low density (ρ ∗ = 0.30) the single chains are more

numerous.

3.3 Results & discussion 45

Figure 3.9: (Color online) Top views from simulations of chains with N = 200 beads in

good-solvent conditions (λ = 0, top) and bad-solvent conditions (λ = 1, bottom), and

with surface-interaction parameter ε∗s = 0.4. (Left) ρ ∗ = 0.81, L = 80σ, Nads = 5200.

(Center) ρ ∗ = 0.49, L = 130σ, Nads = 8200. (Right) ρ ∗ = 0.30, L = 180σ, Nads = 9720.

Figures were prepared using Pymol (http://pymol.sourceforge.net/).

beads per polymer) in bad-solvent conditions, with fixed surface bead density

ρ ∗ = 0.30, and with various surface-interaction parameters ε∗s . Two polymers were

considered clustered if any two beads on different polymers were within a distance

of 1.5σ. Histograms were accumulated from sets of five independent good solvent-

to-bad solvent simulations. With small surface-interaction parameters (ε∗s = 0.4 and

0.6) the distribution shows a monotonic decrease from the peak corresponding to

single chains; with larger parameters (ε∗s = 0.8 and 1.0), the distribution is bimodal,

with a clear delineation between one-chain or two-chain species, and larger clusters.

The bimodal distribution is to be compared qualitatively to that found in AFM ex-

periments, Fig. 3.1(b). A direct, quantitative comparison is not feasible because we

have not considered a specific, coarse-grained molecular model tailored to describe

78.8 kDa linear PB adsorbed on mica. Another factor that might influence the pattern

formation, and that has not been considered in the simulations, is polydispersity

in the lengths of the chains. Nonetheless, we suggest that the general picture, to be

sketched out below, will apply to the real, experimental situation.

3.3 Results & discussion 46

beads in bad-solvent conditions (λ = 1), at a density ρ ∗ = 0.30, and with various

surface-interaction parameters. Each histogram is an average of five independent

simulations of the good solvent-to-bad solvent quenching process. In each case, the

total number of adsorbed beads (as defined in section 3.3.2) is Nads = 10000.

Clearly, the nature of the cluster-size distribution depends on both ρ ∗ and ε∗s .

From simulations of 200-bead polymers at different densities and with different

surface-interaction parameters, we constructed a ‘phase diagram’ indicating whether

the quenched configurations in bad-solvent conditions showed monotonically de-

creasing cluster-size distributions, bimodal distributions, or distributions showing

single peaks; it is shown in Fig. 3.11. In general, low ρ ∗ / low ε∗s favors a monotoni-

cally decreasing cluster distribution. Increasing ε∗s flattens out the polymers on the

surface, while increasing ρ ∗ slightly leads to more overlaps with neighboring chains;

either change leads to more pronounced clustering and a bimodal cluster distribu-

tion. At high values of ρ ∗ and ε∗s , the clustering is extensive and the cluster-size

distribution is strongly peaked (typically at around five chains per cluster).

From these results, we can picture the polymer behavior during the good solvent-

to-bad solvent quench as follows. At low density and with a low surface-interaction

parameter, the polymers are largely isolated from one another and the probability

of forming bead-bead contacts is low because of the large mean separation and the

existence of ‘loops’ and ‘tails’ extending perpendicular to the surface. The polymer

conformations are essentially the same as for an isolated polymer. During the quench,

the majority of polymers simply fold up by themselves; successively smaller propor-

tions of the molecules form dimers, trimers, etc., leading to a monotonic, rapidly

3.4 Conclusions 47

Figure 3.11: (Color online) ‘Phase diagram’ in the plane of surface density ρ ∗ and

surface-interaction parameter ε∗s for polymers with N = 200 beads in bad-solvent

conditions (λ = 1), showing the occurrences of cluster distributions which are either

monotonic decreasing with cluster size, bimodal, or single peaked.

parameter, the polymers are held flat on the surface and hence the chains form many

more contacts with their neighbors. Therefore, during the quench, polymers aggre-

gate with their neighbors and go on to form large clusters. The cluster distribution is

consequently peaked at a relatively large value. At intermediate values of the density

or the surface-interaction parameter, the polymer conformations are not significantly

different from those of isolated polymers, but there are many more opportunities

for forming contacts with neighbors. These factors favor a mixture of the extremal

processes described above, and so give rise to a bimodal cluster distribution.

3.4 Conclusions

spring model to gain insight on the adsorption of linear polymers on to a smooth

surface. The main experimental results we set out to understand are AFM images of

polymers physisorbed from solution on to mica surfaces during a process of rapid

solvent evaporation. We mimicked this process by switching the bead-spring model

interactions from good-solvent to bad-solvent conditions. Of particular interest was

the experimental observation of a bimodal cluster distribution. We have shown that

this feature is favored at low-to-moderate polymer concentrations, and over a broad

range of polymer-surface interaction strengths. At high concentrations, a single-

peaked distribution is observed; at low concentrations, and with weak polymer-

3.4 Conclusions 48

trends observed in simulations have been rationalized in terms of the probable

numbers of contacts between polymers before quenching from good-solvent to

bad-solvent conditions.

We have measured and rationalized the trends in a variety of other properties

including the fraction of monomer units bound to the surface, the height of the

adsorbed polymer film, and the radius of gyration of an adsorbed polymer chain.

An additional, incidental result of this study is the reproduction of an algebraically

decaying density profile within the ‘central regime’, as predicted by de Gennes using

scaling arguments [44]; the simulation results are consistent with the prediction

that the monomer density as a function of the perpendicular distance from the

surface (z) decays like z −4/3 . The existence of the central regime has been confirmed

experimentally [45, 46, 47, 48] and in Monte Carlo simulations of lattice models,

[49, 50, 51] but as far as we’re aware, this has not been demonstrated before in

simulations of an off-lattice model.

Future experimental and simulation work will be focused on the adsorption

and clustering of star polymers on smooth surfaces. In addition, the kinetics of

adsorption and clustering will be explored in detail. For now, we note that two distinct

mechanisms for the self-assembly of adsorbed polymers were identified through

inspecting movies of the simulated quenching process. In the first mechanism,

weakly adsorbed chains first collapse in to individual globules, which then slowly

diffuse over the surface and coalesce. The cluster distribution then appaerently

reaches a steady state on the simulation timescale. This process was more common

with small chains (with N = 50 beads) at low concentrations. The second mechanism

involves the development of contacts between the polymers prior to quenching,

i.e., in good-solvent conditions. Upon quenching, the chains collapse in to one

another, and form more extended structures. Occasionally, we observed a chain

bridging between two others, and causing all three to collapse simultaneously. These

mechanisms were favored by longer chains (N = 200 beads) at high concentrations.

Chapter 4

polymers on surfaces

4.1 Introduction

Star polymers are a special case of branched polymers. They are single molecules

composed by f flexible polymer chains tethered to a core particle, where f is called

the functionality. The average conformation of a star polymer in a dilute solution is

well understood by the standard star blob model, introduced by Daoud and Cotton

[65] and further developed in the work of Birshtein et al. [66, 67]. In the Daoud-

Cotton model, the interior of the star is regarded as a succession of concentric shells

of blobs, see Fig. 4.1. For f = 1 and 2 a star resembles a linear polymer meaning

that for low functionalities, star polymer exhibit highly aspherical conformations

[68, 69]. However, as the functionality is increased, asphericity significantly drops

and the stars become spherical and “stiffer” [70, 71, 72]. A star polymer is therefore a

hybrid between polymer-like objects and colloidal particles bridging two different

domains of physics. A great of interest has been generated recently on how stars

behave in the proximity of a substrate. One such interesting example is how stars

act as depleting agents for colloidal particles, and how they can be used to control

colloidal aggregation to produce desired ordered structures [3, 4, 5]. The situation of

star polymers on flat surfaces is of particular interest not only due to the academic

insights it can provide but also for potential industrial and technological applications.

On the basis of previous theoretical [73, 74, 75] and simulation [76, 77, 78] work

on star polymers on flat surfaces one can categorise the structures as belonging to

one of two categories. In the first case the star polymers are near the adsorption

transition, in which the polymer conformation has (almost) identical statistics when

the star polymer is in the bulk, see Fig. 4.2(a). In the second category the star polymer

49

4.1 Introduction 50

Figure 4.1: (Colour) A scematic illustration of the Daoud-Cotton blob model of a star

polymer.

statistics are greatly affected by polymer’s core being tethered on the surface, see

Fig. 4.2(b). The latter has been studied both on repulsive or marginal surfaces, whose

conformations are called star-brushes and in (strong) attractive surfaces (quasi)

two-dimensional polymer conformations are created. Nevertheless, there is an

intermediate case in which adsorption of the star polymer structure is driven only

by van der Waals interactions (physisorption) on a surface as see Fig. 4.2(c). The

resulting polymer structures are expected to show bulk statistics that is intermediate

between two-dimensional to three-dimensional limiting cases. This intermediate

type of interaction has been been omitted in the literature to date.

linear [1] and star [2] polymers on smooth surfaces using atomic-force microscopy

(AFM). In a typical experiment, a polymer solution was prepared in good-solvent

conditions at concentrations below the critical overlap volume fraction (φ∗ ), resulting

in well-separated chains in solution and hence precluding any strong degree of

structural ordering within the polymer component. Polymer (sub-)monolayers were

formed by exposing a smooth surface (such as highly ordered pyrolytic graphite

or mica) to the polymer solution. The surface was then placed in a good solvent

bath for several hours and extensively rinsed with good solvent to remove any non-

adsorbed polymer. Finally, the samples were dried gently under a stream of nitrogen

and subsequently imaged in air by AFM in tapping mode to investigate the resulting

structures from this good solvent-to-bad solvent ‘quench’. Depending on the polymer

molecular weight, architecture, and concentration, very different surface structures

4.1 Introduction 51

can be obtained.

For the case of linear polymers under certain conditions two distinct types of

cluster were found, giving rise to a bimodal cluster-height distribution. A report of

the experimental investigation is in preparation [1], and a simulation study yielding

insight on the experimental results is in Ref. [79]. Using the same experimental pro-

cedure, we have studied the adsorption of star polybutadiene (PB) on to mica from

toluene. The functionality and concentration of star polymers control a crossover

between ‘polymer’ and ‘soft-colloid’ regimes, being distinguished by characteristic

cluster topologies, sizes, and surface coverages [2]. In this work we aim to repro-

duce, and gain insight on, the transition from star-polymer to soft-colloid regimes

using Langevin dynamics simulations of coarse-grained ‘bead-spring’ models of star

polymers.

In the past, the properties of a star polymer and its crossover between ‘polymer’

and ‘soft-colloid’ regimes has been examined experimentally, theoretically, and with

computer simulations. In particular computer simulations have helped to confirm

the predictions of the Daoud-Cotton model and its scaling concepts [55, 56]. They

have also made possible to obtain the effective interactions between star polymers

and thus further coarse-grain the whole star polymer as one ultra-soft particle [80].

This technique has provided valuable insights on the properties of star polymers,

including the phase diagram [80]. Despite the numerous simulations studies of star

polymers it is still largely unknown how the star polymers behave in the proximity of

an attractive surface. Part of problem, namely, the lack of simulation studies focusing

on this problem, is that the star’s architecture poses a significant challenge for Monte

Carlo (MC), molecular dynamics (MD), and Brownian dynamics (BD) techniques.

For example, lattice Monte Carlo simulations permit only a small number of arms,

less than 20, to be studied, while continuum Monte Carlo runs into problems when

dealing with systems with strong density fluctuations [81]. On the other hand MD and

BD simulations can study the properties of many-arms star polymers, but not yield

information on the number of configuration and thus the calculation of characteristic

exponents in scaling theory.

In this chapter we report a simulation study of adsorbed star-polymer films. We

use Langevin dynamics simulations of coarse-grained ‘bead-spring’ models to gain

insight on the results from polymer adsorption experiments. The outline of the

study is as follows. Firstly, we study the properties of isolated adsorbed polymers

(vanishing surface coverage). For the purposes of comparison with the case of finite

surface coverage, we reconsider specific single-molecule structural properties for

the particular coarse-grained models being employed. Next, we deal with many

polymers on a surface under good-solvent conditions, corresponding to the prelude

4.3 Results & Discussion 52

adsorption; (a) near the adsorption transition; (b) center-adsorbed; (c) physisorbed.

to-bad solvent quench, and its effects on the structure of the polymer film. The

chapter is organized as follows. Section 4.2 contains details of the coarse-grained

polymer model, and the simulation methods. Results for isolated polymers and many

polymers are presented in section 4.3. Section 4.4 concludes the chapter.

Star polymers are composed of f flexible linear chains, which are tethered to a small

central core to form a single molecule. Accordingly, in this work the star polymers are

modelled with f polymer chains, which are attached/bonded to a core particle. Each

chain is composed by N beads of equal size and mass connected by springs. The

size of the core particle may be larger than the size of the beads especially for f > 32,

to allow for the construction of a stable molecule without any bond breaking. We

emphasise that the core size, which was kept as small as possible, is insignificant as

compared to the star’s overall structure. A star of f arms and N beads per arm will be

expressed in the following notation f /N . Further details on the model interactions

and simulation methodology can be found in section 3.2.

We have studied three different situations using Langevin dynamics simulations: (i)

the behavior of an isolated star polymer on a surface in good-solvent conditions

(λ = 0); (ii) the behaviour of an isolated star polymer on a surface in various solvent

qualities (0 < λ ≤ 1); (iii) the behavior of many star polymers on a surface in athermal

solvent conditions (i.e. λ = 0), corresponding to the experimental situation before the

bad-solvent quench; and (iv) many star polymers in bad-solvent conditions (λ = 1)

4.3 Results & Discussion 53

Figure 4.3: (Colour) Screenshots showing isolated stars 4/100, 32/100 and 128/100

from left to right on the top row, and 4/25, 32/25 and 128/25 from left to right on the

bottom row. The interaction strength of surface is ε∗s = 0.6.

studied star polymers with eight different functionalities ( f = 2, 4, 8, 16, 32, 50, 64

and 128), and three different chain lengths (N = 25, 50 and 100), on different surfaces

having surface-energy parameters ε∗s = εs /ε = 0.4, 0.6, 0.8 and 1.0; in the case of

many-polymers, a range of surface coverages was explored.

polymer is reduced, leading to a decrease in entropy. The ‘adsorption transition’ can

be said to be the point where the star polymer is (weakly) physisorbed on the surface,

but the distance of the core from the surface is such that the number of available con-

figurations is not significantly less than the number in bulk. Essentially, the structure

of the star polymer is similar to that in the bulk. Other studies have explored how the

conformation of star polymers with the cores tethered to a surface [76]. In the case of

repulsive or marginal surfaces the core particle being tethered on the surface keeps

the star polymer adsorbed despite the entropic penalties. The configuration of a star

polymer interacting with a marginal surface with a small center-surface distance,

is similar to that of a bulk star polymer with twice as many arms [82]. In the case

of strongly attractive surfaces center-adsorbed star polymers resemble (quasi) two-

dimensional polymers. However, to the best of the knowledge of the authors’ no work

has examined the case of star polymers adsorbed on attractive surfaces but without

the core been tethered on the surface. The resulting star polymer configuration is a

balance between the star’s arms attempt to minimise the energetic contribution of

4.3 Results & Discussion 54

the free-energy arising from polymer-surface interactions and minimise the entropy

penalties for the star being near the surface. Additionally, this biased preference of

monomers towards the surface breaks (alters) the isotropic nature of the star polymer.

In this work we study the surface effects on a physisorbed star polymer and how it

changes its conformation from low to high values of f . We study model parameters

corresponding to the physisorption regime, meaning that the interaction between a

bead and the surface is of order ∼ k B T ; adsorption is favoured because of the large

number of contacts between individual beads and the surface.

All isolated polymer simulations began by placing a star polymer in good-solvent

conditions (λ = 0) and close enough to the surface for adsorption to occur. The

equilibration time needed varied with different values of f , N and ε∗s . Equilibration

was complete within 2 × 105 δt for the smallest molecules and upto 8 × 106 δt for the

largest. Finally, the measured properties were averaged over a production run of

4 × 106 δt .

In Fig. 4.4 we show how the behavior of the average maximum height of the star,

〈h〉, and the average height of the core particle, 〈h core 〉, in different perspectives as a

function of the functionality. In the first perspective we plot the ratios 〈h〉/R c and

〈h core 〉/R c , where R c is the root mean square center-end distance of a star polymer

in bulk,

f

1

(ri − rcore )2 〉1/2 .

X

Rc = 〈 (4.1)

f end points i

The first ratio provides the information on how much the overall structure is affected

by the star’s interaction with the surface, while the second ratio describes the position

of the core particle and how much the star polymer has been deformed by exposing

the core particle on the surface. In both cases, both ratios increase with increasing

f and approach the values 2 and 1, respectively, corresponding to a structure of a

star on a surface which is similar to that in the bulk. Nevertheless, we make some

important observations. The first observation is for the 128/25 star with ε∗s = 0.4, 〈h〉

is slightly above 2R c . This happens because even though the star has a very similar

structure to that in the bulk being quite spherical due to the high values of f , the

few arms having monomer contacts with the surface are stretched to increase the

monomer contacts. The stretching of these arms increase the overall height of the

star polymer. The second point concerns the change in 〈h core 〉/R c in going from

the 2/25 to 4/25 star. The decrease observed is a consequence of a transition from

unbranched to branched architecture. At f = 2 the star is effectively a linear chain

and the position of the core particle can be anywhere between the surface and the

maximum height of the molecule. However, adding one or two more arms creates a

higher density of monomers around the core making it more favorable for it to be

4.3 Results & Discussion 55

Figure 4.4: (Colour) The figures at the top row show the ratio of the maximum height

of the star, 〈h〉, over the end-to-center distance of a star in the bulk, as a function of

f with a surface for various values of ε∗s . At the middle row the figures show the ratio

of the distance of star’s core particle between core particle and the surface, 〈h core 〉,

over the end-to-center distance of a star in the bulk, as a function of f with a surface

for various values of ε∗s . The figures at the bottom row show the ratio between the

height of the core particle, 〈h core 〉, over the maximum average height of the star, 〈h〉.

close to the surface and from the monomer-surface interactions. Nevertheless for

higher values of ε∗s at low f , 〈h core 〉/R c remains low and at the same level, meaning

that the core particle remains close to the surface, see Fig. 4.3 (a) and (d).

In the second perspective, we look at the ratio 〈h core 〉/〈h〉 as function of f . This

quantity describes the relation between the height of the star and the position of its

core. For low values of f the ratio 〈h core 〉/〈h〉 has values slightly less then 0.5, which

means that the core is located slightly lower than the mid-height of the molecule.

The ratio has higher values and closer to 0.5 for more attractive surfaces because

possible tails or large loops from the arms contributing to 〈h〉 diminish. Thus even

though the core particle is close to the surface - as indicated by 〈h core 〉 - the ratio

increases for more attractive surfaces.

Above a threshold value of f the ratio 〈h〉/R c increases significantly, with 〈h core 〉/R c

following the lead but at a different rate. This happens because as f is increased

the number of arms effectively saturates the region near the core of the star, where

the whole structure is essentially flat on the surface. At that point 〈h core 〉/〈h〉 signifi-

cantly drops because any new arms will find the surface too crowded and they will

4.3 Results & Discussion 56

Figure 4.5: (Colour) The figure displays the vertical (continuous line) and parallel

(dashed) components of radius of gyration, 〈RG2 〉⊥ and 〈RG2 〉∥ , for star polymers having

N = 25 in various surfaces. The black line has slope 0.41, corresponding to 〈RG2 〉 ∼

f 0.41 .

become free chains, which greatly contributes to the increase of 〈h〉, see Fig. 4.3(b)

and (e). There are two characteristic conformational features, the first one is that the

adsorbed arms form a disk lying on the surface, and the second feature is that, due

to crowding, some arms become free chains and form a small ‘fountain’, see Fig. 4.3.

At this threshold there is an increase of 〈h core 〉/R c , because the area around the core

particle becomes more crowded and that brings the core particle further away from

the surface.

At higher values of f , the ratios 〈h〉/R c , 〈h core 〉/R c continue to increase towards to

values 1 and 2 respectively. The ratio 〈h core 〉/〈h〉 reaches a minimum and continues

with an increase towards to 0.5. The star polymer essentially starts to regain its bulk

shape and features, and the effects of the surface are reduced. The star polymer star

now starts to resemble a hemispherical, [Fig. 4.3(c)] or spherical droplet [Fig. 4.3(f)].

In other words the star behaves like a soft-colloid, since it is stiff enough to resist the

‘flattening’ effects of the surface. The variation of 〈h core 〉/〈h〉 with ε∗s is in reverse of

what is observed in the ‘linear polymer’ regime. For star polymers, a reduction of the

monomer-surface interaction brings 〈h core 〉/〈h〉 closer to 0.5.

To understand further the behavior of the adsorbed star polymer, we draw atten-

tion to the radius of gyration,

* +

Nt X

Nt

1

〈RG2 〉 = 2

X

(ri − r j ) . (4.2)

N t2 i =1 j =i

4.3 Results & Discussion 57

Figure 4.6: (Colour) The figures of the top row show the ratio of radius of gyration

〈RG2 〉∥ of a star polymer on a surface with diferent ε∗s , over the radius of gyration of a

star polymer in bulk (i.e. no surface), 〈RG2 0 〉∥ = 23 〈RG2 0 〉, as a function of functionality,

f , and for the three diffrent arm lengths N = 25, 50, 100.

where N t = N f +1. It is well known that for star polymers in solution 〈RG2 〉 ∼ N 2ν f 1−ν ,

with ν being the correlation length exponent. In good solvent conditions ν = 3/5,

therefore 〈RG2 〉 ∼ N 1.18 f 0.4 . When a star is away from a surface, 〈RG2 〉 is expected to

scale in a isotropic fashion. But near a surface, we need to consider the dimensions

of the star polymer with respect to the surface. Therefore, we will resolve the vector

ri − r j , in eq. 4.2 in components perpendicular (⊥) and the parallel (∥) to the surface,

and generate the corresponding radii of gyration labelled 〈RG2 〉∥ and 〈RG2 〉⊥ , the sum

of which equals 〈RG2 〉. As seen in Fig. 4.5 for the case N = 25, the components of

〈RG2 〉 deviate significantly for low functionality and for higher values of the surface

interaction strength, ε∗s . At very high values of f , the star is “stiff” enough to maintain

the spherical shape it had away from the surface. Thus in the proximity of a surface

isotropy breaks down with the collapse of scaling with f , that is 〈RG2 〉 is no longer

∼ f 0.4 , see Fig. 4.5. The strong interaction with the surface significantly alters the

structure of a low- or mid-functionality adsorbed star.

In Fig. 4.6 we plot the ratio of the parallel component of the radius of gyration

of an isolated star on the surface over its value in the bulk, 〈RG2 〉∥ /〈RG2 0 〉∥ . The ratio

essentially describes the degree of the effects on the shape/size of the star caused

by the surface. The ratio has values higher than 1 since the polymer spreads on the

lateral directions of the surface. As seen above a low functionalities the star has a

two-dimensional conformation and for high functionalities the star polymer regains

its bulk statistics. A star polymer having a two-dimensional conformation means that

we can use the scaling arguments for two-dimensional stars that is 〈RG2 〉∥ ∼ f 1−3/4 ,

4.3 Results & Discussion 58

where we have used the value of characteristic exponent, ν = 3/4, in two dimentions.

This leads to the prediction that

However, the results in Fig. 4.6 do not agree with eq. 4.3. We interpret the results

as follows. At f = 2 (equivalent of linear chain) the ratio is > 1 meaning that the

star has spread out on the surface compared with that away from the surface. As

f increases a more complex image emerges. Instead of following eq. 4.3 there is a

significant decrease of the ratio meaning for example that a 4/50 star is not affected

as much as a 2/50 star on a surface. This becomes more pronounced for stronger

surface interactions. The interpretation for this behavior is that each arm of the star

polymer behaves as a linear chain. For low values of f each arm on the surface is

well separated from the rest of the arms of the star and hence there is little or no

interaction between the arms. In other words each arm is affected by the surface in

the same way as in the isolated chain of the same length. This interpretation explains

the ratio’s values for 4/50 and 4/100 stars being at the same level with the values of

2/25 and 4/50 stars, respectively. Overall, it indicates that a star of low functionality

and with long arms on a strong attractive surface resembles a network of linear

chains, with the core particle being its hub (or contact point). However, this behavior

is expected to be a finite effect. As obeserved in Fig. 4.6 the depth of the minimum is

smaller for N = 100 than for N = 50. Hence, for longer arms the depth would become

smaller, and the behaviour would be better described by eq. 4.3. As f increases there

is an increase of 〈RG2 〉∥ /〈RG2 0 〉∥ , because the new arms start to interact and to compete

for space (crowding effect), which leads to the star stretching out more on the surface.

As seen in Fig. 4.3(a) and (d), the star resembles a two-dimensional object, thus new

arms will have more pronounced effects on the structure of the star polymer. For

high values of f the ratio descreases and approaches to 1, because the star starts to

distance itself from the surface (also see Fig. 4.4) and regains the bulk characteristics

and shape.

In Fig. 4.6 we show the average bound fraction 〈Φa 〉 as a function of f , where

the bound fraction is defined as the fraction of beads within the interaction range

of the surface defined using the distance-based criterion z ≤ 1.2σ. For details see

subsection 3.3.1. At low f the behavior of 〈Φa 〉 remains at the same level for the case

of linear chains ( f = 2), since all arms are adsorbed on the surface and essentially

each behaves like a linear chain. Having more arms, will create crowding effects and

after a certain threshold the star will have arms not being adsorbed, thus reducing

the bound fraction.

4.3 Results & Discussion 59

Figure 4.7: (Colour) Snapshots of 64/100 star polymer. The top snapshot shows

the star polymer conformation in the bulk. The mid and bottom snapshots display

the topview and sideview, respectively, conformation of the star on a surface with

ε∗s = 1.0 showing how much it has spread on the surface.

at its shape. There many ways to characterize the shape of an arbitary object; in

particular for polymers one can gain valuable insights by looking at the shapes of

random walk, for an overview conserning star polymers see Ref. [83] In this work we

will use the characterization developed by Runidick et al.[84] in which the asphericity,

A d , is defined as

Pd D 2 2

E

j >i (L i

− L j

)

Ad = (4.4)

2 2

D¡P ¢ E

d

(d − 1) L

i =1 i

where L i are the principal components of the squared radius of gyration, and d is

the number of the principal components. The quantity A d has 0 as its lower bound

4.3 Results & Discussion 60

Figure 4.8: (Colour) The figure displays the degree of asphericity, for N = 25, 50 and

100 as a function of f for different values of ε∗s . In the top row the ratio of A x y z /A ∗x y z

and in bottom row the A xz are plotted.

and it is achieved when the object has a spherical shape, and has upper bound of

1 when it is extended in one dimension (rod-like). In Fig. 4.8 we plot A x y z (three-

dimensional asphericity) of a star in the bulk and on a surface for different surface

energy interactions. It provides information on how much the surface has affected a

star’s shape. In previous studies it was found that in the bulk the asphericity drops as

∼ f −1 [83], which is in agreement with our results in the bulk. However, the results

for stars on the surface show a more complex picture. For this reason in Fig. 4.8 we

also plot the A x y , which is the two-dimensional asphericity with the beads’ positions

projected on to the surface. The quantity A x y has a weak power-law dependence that

scales with ∼ f −1.4 , which is different from that found in the bulk ∼ f −1 . Moreover,

in Fig. 4.8 we plot the asphericity in the x, z directions, A xz that is one parallel and

one perpendicular to the surface. The values of A xz provide a more clear description

of an adsorbed star measuring the degree of flatness. The results show that for high

values of ε∗s and low values of f , A xz has values near 1 meaning that the profile of the

star spans only the dimensions parallel to the surface. In other words the star lies

flat on the surface. Note that the longer the arms are, the higher the functionality

is required to start regaining its bulk shape; in the case of N = 100 the threshold is

f = 32. That is because longer arms have more monomer contacts with the surface

and with their energetic contributions the star adopts configurations with higher

4.3 Results & Discussion 61

Figure 4.9: (Colour) The figures display the density profiles of adsorbed stars of

f /100 on surfaces with ε∗s = 0.4 (top) and ε∗s = 1.0 (bottom).

entropic costs. Thus one can adjust the strength of the surface interaction and the

length of the arms of a star and convert a high f star into a (nearly) two-dimensional

object, see Fig. 4.7. A more detailed investigation of the transition of an adsorbed

star between being a two- and three-dimensional regimes is in preparation.

In Fig. 4.9 we plot the probability density p(z) of finding a bead at a perpendicular

distance z from the surface for ε∗s = 0.4 and ε∗s = 1.0 for adsorbed stars with f /100

stars. In all cases there is either a local minimum or a point of inflection in p(z) at

z ' 1.2σ, and the position of these features was taken as a distance-based criterion for

assessing whether a particular bead is ‘bound’ to the surface or not. Note that from

p

6

eq. 3.7 the minimum bead-surface potential energy is located at z/σ = 2/5 ' 0.86;

this distance corresponds to the positions of the primary peaks in p(z). The first

two peaks observed in Fig. 4.9 arise from two ordered bead layers on the surface. A

third peak is observed in the case of ε∗s = 0.4 at f = 64 and 128; it corresponds to

the position of the star’s core particle. Additionally, for z > h core the density profile

p(z) ∼ z −4/3 , which corresponds to the Daoud-Cotton picture [65], in which the bulk

interior of a star in good-solvent conditions and with suffiently long arms possesses a

region in which the monomer density profile follows a power law as a function of the

distance from the star’s center. In other words, the arms of high-functionality stars

remote from weakly attractive surfaces remain unpertubed by adsorption effects.

From the results presented here three different structures have been identified,

and we use the following terminology. The linear regime corresponds to the case

where the arms do not interact significatly with one another, and so the star polymer

4.3 Results & Discussion 62

Figure 4.10: (Colour) The figure display the different regimes of a N / f star polymer

on a surface with ε∗s = 1.0.

acts like a network of loosely-coupled linear chains. The colloidal regime applies

when the star resembles an ulta-soft colloid. The star regime is intermediate be-

tween the linear and colloidal regimes. Within the star regime, a star polymer on

a surface may only extend in the lateral directions and thus it behaves like a two

dimentional object. We identify the boundaries between the linear and star regimes

using the minimum occuring of 〈RG2 〉∥ /〈RG2 0 〉∥ in Fig. 4.6. The transition between

a two-dimensional star and three-dimensional star occurs at the point just before

〈h core 〉/〈h〉 significantly decreses, after which at least one arm is desorbed from the

surface. Finally, the boundary between the star and colloidal regimes is identified

using the minimum occuring in 〈h core 〉/〈h〉. These regimes mentioned above are

plotted in a ‘phase diagram’ as shown in Fig. 4.10.

In this section we focus on the effects of the solvent on the structure of an isolated

adsorbed star polymer. Once the star polymer had adsorbed in athermal solvent

(λ = 0) followed the following procedure described in the previous section, the

solvent quality was adjusted by changing λ to the desired value. The molecule

was equilibrated for around 2 × 106 δt , and then properties were measured over a

production run of 4 × 106 δt . Note that if a change in solvent quality was more than

0.4 = λnew −λold then intermediate equilibration cycles of 106 δt were performed after

increments of ∆λ = 0.1 until the desired value was reached. The solvent parameter

values we have explored, apart from λ = 0 corresponding to athermal solvent, are

4.3 Results & Discussion 63

Figure 4.11: (Colour) The diagrams display the behavior height an adsorbed star

polymer f /50 (top row) and its core particle (bottom row). For comparison the

quantities are presented for different monomer-surface interactions, ε∗s = 0.4 (left

column) and ε∗s = 1.0 (right column).

Changing the solvent quality means that the monomer-monomer interactions

become less repulsive and the crowding effects are reduced. This has an immediate

result; the star polymer shrinks in size. In Fig. 4.11 this effect is observed by measur-

ing 〈h〉 and 〈h core 〉. Both quantities are reduced by descreasing the solvent quality.

Nevertheless, a deviation from this behavior has been observed when entering in

to the bad-solvent regime for small values of ε∗s , i.e. ε∗s = 0.4. The height of the star

polymer is getting smaller, while the height of the core particle, 〈h core 〉 is slightly

increased. This is more pronouced for higher values of f . The interpretation of this

behavior is that the monomer-monomer interactions become significant compared

to the monomer-surface interactions; the monomers adopt a more spherical config-

uration in an attempt to minimize the surface contact with the bad solvent. Adopting

a more spherical configuration pushes the position of the core particle towards the

center of the star and away from the exterior surface of the star.

In Fig.4.12 we present results for 〈RG2 〉∥ and Φa . As described above when the

solvent quality is lowered the star polymer shrinks in size. The behavior of 〈RG2 〉∥

agrees with this description for both monomer-surface interactions being considered

(ε∗s = 0.4 and 1.0). On the other hand the bound fraction differs by exhibiting the

following interesting behavior. Φa remains at the same level as the solvent quality

changes from good-to-ideal solvent conditions. A small decrease of Φa is observed

when the star polymer enters into bad-solvent conditions. For ε∗s = 1.0 the entrance

4.3 Results & Discussion 64

Figure 4.12: (Colour) The diagrams display the behavior of the 〈RG2 〉∥ of an isolated

adsorbed star polymer f /50 (top row) and its bound fraction 〈Φ〉 (bottom row). For

comparison the quantities are presented for ε∗s = 0.4 (left column) and ε∗s = 1.0 (right

column).

into bad-solvent conditions Φa has a more significant decrease than for ε∗s = 0.4.

In Fig. 4.13 we summarize the effects of the solvent on the star’s conformation.

As the solvent quality shifts from good to ideal the star structure “shrinks” in the

sense that the crowding effects are not as pronouced as in the case of athermal

solvent conditions, compare Fig. 4.13(a) and (b). The maximum height of the star

and the height of the core are both reduced. Additionally, the bound fraction remains

roughly at the same level. Nevertheless, as we shift from ideal-solvent to bad-solvent

conditions two different behaviors are observed. In the first one the star polymer

attemts to minimise its surface contact with the bad solvent with a hemishperical

or almost spherical shape (droplet). The height of the star is further reduced but

the height of the core is increased because it moves closer to the center of the star’s

spherical shape, see Fig. 4.13(c). The dominant factor behind this behavior is that

the monomer-monomer interactions become energetically more significant than the

monomer-surface interactions. If the monomer-surface interactions still dominate

the energetic contributions to the free energy then the overall shape of the star will

be similar to that shown in Fig. 4.13(d), where both the heights of the star and its core

are reduced.

Additionally, a strong monomer-surface interactio can have drastic effects on

the structure in the lateral directions. Fig. 4.14 shows a comparison between star

configurations on surfaces with ε∗s = 0.4 and ε∗s = 1.0, these differences can be ob-

served. In the case of ε∗s = 0.4 the collapsed polymer has a nearly globular shape, but

4.3 Results & Discussion 65

polymer on an attractive surface; (a) good solvent conditions; (b) near θ-solvent

conditions; (c) bad solvent conditions and the monomer-monomer interaction is

stronger than the monomer-surface interaction; (d) bad solvent conditions and the

monomer-monomer interaction is weaker than the monomer-surface interaction.

in the case of ε∗s = 1.0 the star has an irregular/asymmetrical shape. Nevertheless,

the latter converges to a more regular/symmetrical shape for high values of f , i.e.

f ≥ 32. This is an important observation because with strongly interacting surface

one can manipulate the shape of the star polymer in the bad solvent conditions.

In good-solvent conditions, polymers experience purely repulsive effective mutual

interactions. Appropriate simulations with λ = 0 were initiated by preparing config-

urations with many polymers on a surface, and equilibrating for around 5 × 106 δt .

Due to computational constraints we were restricted by the number of molecules,

N p = 20 000/ f N , where 20 000 is the total number of beads used in all simulations

(without counting the core particles). Increasing the number of arms, the number of

molecules decreases thus making simulations of even modest values of f computa-

tionally expensive. For this reason in this study we investigated only the following

star polymers 2/25, 4/25, 8/25 and 16/25. Following equilibration, we performed a

production run of 4 × 106 δt .

The adsorption is measured by the equilibrium surface bead density, defined in

terms of the number of beads Nads belonging to those stars with at least one bead-

surface contact, defined using the distance-based criterion z ≤ 1.2σ. Note that Nads

is, in general, greater than the number of beads actually bound to the surface. The

reduced surface bead density is ρ ∗ = Nads σ2 /L 2 . Furthermore, in the adsorbed state

we define an effective intramolecular two-dimensional density of an isolated star ρ ∗

as

N

ρ iso = (4.5)

A iso

4.3 Results & Discussion 66

solvent conditions on surfaces from a top-view perspective. The top row is for ε∗s = 0.4

and the bottom row is for ε∗s = 1.0. From left to the right the columns display the

configurations of stars with 8/100, 16/100 and 32/100, respectively.

where A iso is the area covered on the surface by the adsorbed polymer, given by

A iso = π〈RG2 〉∥ , where 〈RG2 〉∥ is the parallel component of the radius of gyration of

an isolated adsorbed polymer. ρ iso can also be used to define the critical overlap

concentration between star polymers. The dilute regime corresponds to ρ ∗ ¿ ρ iso ;

in this regime the individual stars are almost independent and nearly unperturbed.

For ρ ∗ > ρ iso , the excluded volume constrains lead to interaction effects between

adsorbed star polymers. Now, we introduce the scaling variable

ρ∗

γ= (4.6)

ρ iso

number of stars on surfaces of various dimensions, we simulated surface densities,

up to that corresponding to the critical overlap concetration. In other words, the sim-

ulations were perfomed at surface concetrations (within the film) approaching the

semidilute regime. During equilibration runs near the critical overlap concentration,

some of the polymers were seen to desorb as the polymer film approached the steady

state.

In Fig. 4.15 we present the behavior of 〈h〉 and 〈h core 〉 with varying surface density.

For very small surface densities the star polymers are dispersed on the surface, and so

we expect identical results with those of an isolated star polymer. Nevertheless, the

4.3 Results & Discussion 67

Figure 4.15: (Colour) The diagrams display the behavior of the average maximum

height of adsorbed stars f /25 (top row) and their core particles (bottom row). For

comparison the quantities are presented for for different monomer-surface interac-

tions, ε∗s = 0.4 (left column) and ε∗s = 1.0 (right column).

surface density critical value beyond, which the polymer conformation significantly

changes, varies with functionality and the monomer-surface interaction. Under

weak-adsorption conditions (ε∗s = 0.4) only small values of γ could be addresses due

to desorption, which reduced the number of star polymers on the surface. For the

case of linear chains ( f = 2) one expects the critical density threshold to be much

higher than for f > 2. This is confirmed for strong adsorption conditions. Under

weak adsorption conditions the chain conformation is more sensitive to changes in

density. A possible interpretation is that the adsorbed chains can overlap with other

chains without entropic restrictions by adopting configurations that contain long

tails or loops.

This effect can be seen in Fig. 4.15, which shows that the height of the core particle

is more sensitive to changes in density for lower functionality star polymers. The

rate of change of 〈h core 〉 with density for f = 2 is much higher than that for f = 4 or

8, while in the case of f = 16 the rate remain negligible. For f > 2 the values of 〈h〉

remain at the same level within the range of 0 ≤ γ < 0.2; beyond this range there is an

increase. That is because at γ ≈ 0.2 the star polymers start to feel the presence of their

neighbouring star polymers. Similar behavior is observed under strong-adsorption

conditions (ε∗s = 1.0). The difference is that the density threshold, for the star polymer

statistics to change is at γ ≈ 0.5. In the strong monomer-surface interaction case,

〈h〉 starts to increase earlier than 〈h core 〉, due to crowding monomer effects on the

surface. This situation arises by the star polymer shrinking in the lateral directions

4.3 Results & Discussion 68

Figure 4.16: (Colour) The diagrams display the behavior of the ratio of the parallel

over the vertical component of the radius of gyration, 〈RG2 〉∥ /〈RG2 〉⊥ , of adsorbed stars

f /25 (top row) and their bound fraction (bottom row). For comparison the quantities

are presented for for different monomer-surface interactions, ε∗s = 0.4 (left column)

and ε∗s = 1.0 (right column).

and then, if the crowding effects persist, some arms desorb from the surface. The

reasoning is similar to the case of isolated star of varying functionality.

In Fig. 4.16 we display the behavior of the ratio of the parallel and the perpendic-

ular components of the radius of gyration, 〈RG2 〉∥ /〈RG2 〉⊥ , and the bound fraction, Φa .

The results show that under weak-adsorption conditions small structural changes

occur with a small decrease of 〈RG2 〉∥ /〈RG2 〉⊥ by about less that 20%. Even though

γ < 0.6 meaning there is still enough space on the surface for more star polymers

to be adsorbed, the star polymers do experience small configurational changes due

to their interactions with neighbours. One interpretation for this is that the star

polymers diffuse on the surface and eventualy they will “collide” with each other.

Once such a collision occurs there would be a degree of interpenetration, which will

affect the overall conformation of both colliding stars. Additionally, the functionality

plays an important role both on the degree of interpenetration and resulting confor-

mational changes. For example, a star polymer of f = 2 will diffuse more easily on the

surface than a star of f > 2 (keeping the arm length fixed). Additionally, desorption of

one of the arms will be more significant in a star of lower functionality. This explains

why the conformation of a star f = 2 is most sensitive to density changes.

For strong adsorption more significant changes occur, because the star polymer

extend further in the lateral directions. The star polymers possess higher values of

〈RG2 〉∥ - as shown in Fig. 4.5 - and thus they are more likely to overlap with other

4.3 Results & Discussion 69

Figure 4.17: (Colour) Top view from simulations of star polymers on an attractive

surface ε∗s = 0.4 in good-solvent conditions (λ = 0, top) and bad-solvent conditions

(λ = 1.0, bottom). All stars in the figure have N = 25 and from left to right ( f , γ) =

(2, 0.107), (4, 0.186), (8, 0.466) and (16, 0.750), respectively.

star polymers. Another key difference with the weak adsorption case is that the

configuration of the star polymer will be in the linear- or 2D-star regime. The arms

will be flatter on the surface, greatly reducing the opportunities for interpenetration.

Another factor that controls the degree of interpenetration is functionality. As func-

tionality increases both the ratio 〈RG2 〉∥ /〈RG2 〉⊥ and Φa drop, as shown in Fig. 4.16.

This is no surprise since functionality controls the “stiffness” of the stars.

here, is the quench from good-solvent to bad-solvent conditions, corresponding

to rinsing with solvent and then drying in nitrogen/air. As described in chapter 3

for the case of linear chains we mimic this step in our simulations by starting sim-

ulations from well-equilibrated configurations with λ = 0 (good solvent) and then

instantaneously switching to λ = 1 (bad solvent). We then re-equilibrate the system

for 5 × 106 δt , during which the system was seen to reach an apparent steady state.

As discussed in the case of linear polymers in chapter 3 the surface structures are

apparently stable for a long period of time. Star polymers diffuse less as functionality

is increased, and so in our simulations the structures observed are apparently static

on the timescales accessed in the simulations.

In Fig. 4.17 we show examples of equilibrated simulation configurations before

(λ = 0) and after (λ = 1) the quench, for systems of polymers each made up from

4.4 Conclusions 70

N = 25 beads per arm, and with surface-interaction parameter ε∗s = 0.4, with dif-

ferent functionalities, and at various surface densities. In good-solvent conditions

the polymers are in extended conformations, but in bad-solvent conditions they

collapse to optimise the attractive bead-bead interactions. Functionality plays a

significant role in determining the degree of clustering between the polymers. As

seen in Fig. 4.17 even at a low densities, a low functionalities of f = 2 or 4 leads to

clusters composed of 4 or even 6 polymers, while for f = 16 one has to reach higher

values of surface coverage (γ ≈ 0.750) to observe a cluster composed of 3 polymers.

The interpenetration of this is based on ‘stiffness’ of the star polymer before the

quench. For low functionalities the star polymers diffuse in the lateral directions

and once they ‘collide’ they easily interpenetrate with one another. Once the quench

occurs some of the polymers will overlap each other or they are very close together

meaning that is highly probable they will form clusters. However, for higher values of

f the degree of interpenetration between polymers is greatly reduced. Therefore high

functionality stars, prior the quench, position themselves in such a way to minimise

the degree of overlapping with each other. Hence, after the quench the star polymers

tend to form isolated collapsed globules or small clusters. To observe clustering of

high functionality star polymers, one has to significantly increase the surface density

resulting an increase to the degree of polymer overlap. Another important factor that

contributes is that the functionality greatly reduces the mobility (diffusion) of the

star polymers on the surface. This means that the positions of the center of mass

of the star polymers on the surface before and after the quench will be roughly the

same. Exceptions to this would occur if a star polymer has many bead contacts or

entanglements with a neighbouring star polymer prior the quench.

4.4 Conclusions

spring models to gain insight on the adsorption of star polymers on to a smooth

surface. Our motivation was driven by experimental AFM images of polymers ph-

ysisorbed from solution on to a mica surfaces during a process of rapid solvent

evaporation. The process was mimicked by switching the coarse-grained model in-

teractions between those appropiate for good-solvent or bad-solvent conditions. Of

particular interest was the experimental observation of the crossover from polymer-

like to colloidal behavior. Initially, we examined how the crossover manifests itself in

an isolated adsorbed polymer. Three parameters were identified being important

for controling the structure of the star on the surface: f , N and ε∗s . By exploring the

4.4 Conclusions 71

parameter space we mapped the different behaviors onto a ‘phase diagram’. Addi-

tionally, our methodology allowed us to change the solvent quality by tuning one

parameter, λ, thus gaining valuable insights on the conformational behavior of the

adsorbed star in different solvent conditions. Finally, we explored systems of many

adsorbed star polymers under good-solvent conditions and the corresponding pat-

terns they create once quenched into bad-solvent conditions. The observed patterns

were rationalised in terms of the probable interpenetration between polymers before

quenching from good-solvent to bad-solvent conditions.

Chapter 5

hard disk fluid

5.1 Introduction

For over four decades the science of complex networks was based on results for

random graphs, obtained by Erdös and Rényi in 1959 [85]. Recent analyses reveal that

the topologies of most large, real networks deviate from those of random networks

[86]. Examples of real networks include the World Wide Web [87, 88], the Internet

[89], metabolic systems and protein-protein interactions [90, 91, 92], sexual contacts

[93], collaborations between scientists [7] and between movie actors [94], and scoring

totals of Brazilian football players [95]. The most significant deviations from random-

network behavior appear in the tail of the degree distribution p k , that is the probability

of a given vertex possessing connections to k other vertices. In random networks

p k is Poissonian, while p k for many real networks varies asymptotically like k −γ ,

where γ is a positive exponent [94]. In qualitative terms, a scale-free network is

characterized by a small number of highly-connected vertices called ‘hubs’; for a

general introduction, see Ref. [96]. Many authors identify scale-free networks as

those possessing power-law degree distributions, but as discussed fully in Ref. [97],

this is a necessary but not sufficient condition. In what follows, we will be largely

concerned with power-law degree distributions, and the term ‘scale free’ will be

employed; but it should be borne in mind that, strictly speaking, other properties are

required for the network to be termed scale free [97].

It has been suggested – but not yet confirmed – that growth and preferential

attachment are possible mechanisms by which real-world, scale-free networks can

emerge [86]; the addition of nodes to networks may capture the intrinsic evolutionary

behavior of some real-world examples. Dynamical models of the growth of cellular

72

5.1 Introduction 73

Figure 5.1: An example of an Apollonian gasket. The image was taken from Wikipedia

(http://en.wikipedia.org/wiki/Apollonian_gasket).

networks – including the effects of cell division and disappearance – can also give rise

to scale-free neighbor networks [98, 99]. Stochastic models have been proposed and

studied which can successfully reproduce certain features of real networks [94, 100],

however there are some features which cannot yet be reproduced. An example is

the assortativity of a network [101], which measures the extent to which vertices

with equal degree link with one another: in assortative networks, vertices with equal

degree are linked to one another more frequently than in a random network; in

disassortative networks, vertices with equal degree are linked to one another less

frequently than in a random network. To quote Newman, “An interesting observa-

tion is that essentially all social networks measured appear to be assortative, but

other types of networks (information networks, technological networks, biological

networks) appear to be disassortative. It is not clear what the explanation for this

result is, or even if there is any one single explanation. (Probably there is not.)" [102].

Nonetheless, it has been shown that assortativity and clustering (which means more

linked vertices share a common neighbor than would be expected by chance) can

arise if social networks are divided in to distinct groups or communities [103].

Scale-free networks are rare in real and model condensed-matter systems. One

important example is the scale-free contact network in the Apollonian packing of cir-

cles [104], in which the circle diameter distribution decays according to a power-law

with an exponent of about 1.3 [105]. The mechanical, percolation, and conductive

properties of materials possessing Apollonian-packing networks have been shown

to exhibit unusual dependences on the number of vertices [104]. In Fig. 5.1 displays

5.1 Introduction 74

an example of Apollonian disks. Such models may describe the distribution of force

in granular materials such as concrete, and networks of voids in porous media. The

percolation properties of such networks [104, 106] will clearly be of relevance to the

characteristics of certain porous media. The Apollonian packing has also been iden-

tified as a model for networks of connected minima on potential-energy surfaces of

atomic clusters [107]; the transition states (edges) between minima (vertices) exhibit

a scale-free distribution [108, 109, 110].

This chapter is concerned with the possibility of generating a simple scale-free

network in molecular simulations of a simple fluid. The model consists of poly-

disperse hard disks in two dimensions, in which the distribution of disk diameters

(σ) varies asymptotically like σ−α , where α is a positive exponent. The equilibrium

structure of the fluid phase is determined using Monte Carlo (MC) simulations, con-

ducted at constant pressure (strictly tension). Naïvely one might expect the number

of neighbors of a given disk to be proportional to the circumference and hence σ.

Due to the broad distribution of particle diameters, neighbors are identified using the

radical tessellation [111] proposed as a suitable alternative to the Voronoï construc-

tion which is usually applied to one-component systems. If the tail of the diameter

distribution follows a power law, then perhaps the neighbor distribution will vary in

a similar way. Identifying particles with vertices, and the separation vectors between

neighboring disks as edges, this situation could give rise to a scale-free neighbor net-

work, with the large particles playing the role of the hubs. In this section it is shown

that: (i) there are indications that the neighbor network in the polydisperse hard-disk

fluid possesses a power-law degree distribution, which is one of the properties of a

scale-free network; and (ii) the neighbor network is highly assortative, i.e., vertices

with equal degree are directly connected more often than in a random network. This

is an interesting situation, because the equilibrium properties of the fluid are static,

and hence dynamic mechanisms of scale-free network formation, e.g., preferential

attachment, are not applicable.

Polydisperse hard-disk fluids are of considerable intrinsic interest, for example as

models of colloidal monolayers, froths and foams, and packing and segregation in

granular materials. The packing of binary mixtures of hard disks has been studied

extensively and is shown to be highly non-trivial [112, 113, 114]; for instance, the dis-

tributions of neighbors and of cell area (as obtained from radical tessellations) show

distinct contributions arising separately from the small and large disks, indicating

the clustering of disks with equal size. Fluids with more than two components have

received far less attention, although it has been noted than in some respects the statis-

tics of the radical tesselations may conform to those of a random tesselation [112].

Specifically, it has been found that fluids of disks with a linear, decreasing diameter

5.2 Model and methods 75

distribution obey the Aboav-Weaire law, which states that the average total number of

sides of the cells (determined by tesselation) neighboring a cell with k sides increases

linearly with k [115, 116, 117]. It has been shown that the Aboav-Weaire law is a con-

sequence of a maximum-entropy (ME) principle [118, 119], the argument being that

it can be written as a linear superposition of two basic constraints, these being that

P P

k p k = 1, and for a two-dimensional tesselation k kp k = 6 (Euler); therefore, the

Aboav-Weaire law provides no new constraints and so it leads to the ME distribution.

It might therefore be anticipated that in the polydisperse hard-disk fluid considered

in this work, a maximum-entropy distribution may provide an adequate description

of the nearest-neighbor network measured in simulations.

From the viewpoint of liquid-state physics, it is interesting to construct expres-

sions for the equations of state of highly polydisperse fluids. In statistical mechanical

terms, this may stimulate progress in understanding dense, complex fluids, such as

the mixtures of macromolecules found inside living cells [120]. Therefore, a simple

equation of state is derived using scaled-particle theory (SPT) [121], and compared

with (essentially exact) results from MC simulations.

This chapter is organized as follows. Section 5.2 details the hard-disk model,

summarizes the MC and SPT approaches to be employed, and presents the derivation

of an ME estimate of p k . The results are reported and discussed in Section 5.3, and

Section 5.4 concludes the paper.

The model consists of N hard disks confined to a square cell of area A. The disk-

diameter distribution is given by

C

p(σ) = (5.1)

1 + (σ/σ0 )α

malization constant. This particular choice of distribution was chosen so that there

is a finite number of small particles (important for the simulations), and because it

rapidly approaches the asymptotic form p(σ) ∼ σ−α . The value of α should be small

enough to give a long tail to the distribution, but large enough such that the first few

moments are well defined. The second moment is required to exist so that one can

define a packing fraction, given by η = πρ〈σ2 〉/4, where the angled brackets denote

an average over p(σ), and ρ = N /A is the number density. The second moment is

finite only for α > 3, so present work is focused on the distribution with α = 4, the

5.2 Model and methods 76

a log-log scale to highlight the asymptotic σ−4 behavior.

p

2 2

C = , (5.2)

πσ0

σ0

〈σ〉 = p , (5.3)

2

2 2

〈σ 〉 = σ0 . (5.4)

p(σ) is plotted in Fig. 5.2 on a log-log scale to show the crossover to a power-law

decay for σ & σ0 .

hard disks were performed in a square cell with periodic conditions applied [9].

The hard-disk system is athermal and so the reduced pressure employed in the

simulations is defined as P ∗ = P σ20 /k B T , where k B is Boltzmann’s constant and T is

the temperature. Ten different simulation configurations were used, each consisting

of particles with diameters drawn randomly from p(σ). Results obtained with the

ten configurations at a given reduced pressure were averaged.

On average, one MC cycle consisted of N attempted displacements of randomly

selected disks, N A trial displacements of ln A [9], and one cluster move, the latter

implemented using Dress and Krauth’s cluster algorithm [122] with reflection rather

5.2 Model and methods 77

Figure 5.3: A scematic illustration of the cluster move algorithm. (a) Initial config-

uration. The black dot it the random pivot point; (b) the randomly chosen particle

is reflected with respect to the pivot point; (c) particles overlaping with the new

position of particle 1 are reflected; (d) the final configuration.

than rotation [123]. The cluster moves are vital for equilibration, because the dis-

placements of large disks are severely hampered by the large number of neighboring

small disks. In a few words, each cluster move begins with the selection of a random

pivot point within the simulation box. A disk is chosen at random, and is subjected

to a point reflection about the pivot. Any disks that overlap with the reflected disk are

themselves reflected, and this process is repeated until there are no more overlaps,

see Fig. 5.3. This algorithm works particularly well at low to intermediate densities,

but it fails at high densities due to all of the particles being part of the same ‘cluster’

[123]. For each state point the equilibration phase consisted of about 105 MC cycles

with N A = 1-20, depending on the pressure. Production runs consisted of a further

105 MC cycles. At intervals of 10 MC cycles, radical tessellations [111] were com-

puted and the network statistics were incremented. In the standard two-dimensional

Voronoï construction, the edges of the tessellation are perpendicular bisectors of

the lines joining the centers of neighboring disks. When neighboring disks are of

very different size, the bisectors may intersect with the larger disks. In the radical

tessellation, this problem is avoided by forming edges with the loci of points from

5.2 Model and methods 78

The derivation of a simple equation of state from SPT [121] is now summarized. To

begin, the excess chemical potential (in units of k B T = β−1 ) of a scaled disk with

diameter λσ is given in the limit of small λ by the Widom insertion formula [124]

· Z ¸

ex 0 0 0

βµ (λσ) ≈ − ln 1 − ρ p(σ )v(λσ, σ )dσ (5.5)

where v(λσ, σ0 ) is the excluded ‘volume’ of two hard disks with diameters λσ and σ0 :

π

v(λσ, σ0 ) = (λσ + σ0 )2 . (5.6)

4

The combination of Eqs. (5.5) and (5.6) can be expanded about λ = 0 up to a linear

term. An additional λ2 term is chosen to yield the correct limit when λ → ∞; in the

case of a scaled disk with area πλ2 σ2 /4 the reversible work of expanding the particle

against the macroscopic pressure is πP λ2 σ2 /4. The result of these manipulations is

an approximation to the excess chemical potential at fixed density and temperature:

πρσ〈σ0 〉 π

βµex (λσ) ' − ln (1 − η) + λ + βP σ2 λ2 . (5.7)

2(1 − η) 4

To obtain an equation of state, appeal is made to a relation for hard particles derived

by Smith and Labík [125, 126]. Consider the quantity Y defined by the derivative

of βµex with respect to λ at λ = 1, averaged over the diameter distribution of the

inserted particle:

dβµex (σ) ¯¯

Z ¯

Y = p(σ) dσ. (5.8)

dλ ¯

λ=1

a full-sized particle against the surrounding fluid. This is given by [127]

∂v(λσ, σ0 ) ¯¯

Z ¯

0 0

Y = ρ p(σ)p(σ )g (σ, σ ) dσdσ0

∂λ ¯

λ=1

1

Z

= πρ p(σ)p(σ0 )g (σ, σ0 )σ(σ + σ0 )dσdσ0 (5.9)

2

where g (σ, σ0 ) is the partial pair correlation function for particles with diameters

σ and σ0 , at contact. The right-hand side of this equation is simply related to the

compressibility factor, Z = βP /ρ, which is easily obtained from the virial equation

5.2 Model and methods 79

[128]:

1

Z Z

Z = 1 + πρ p(σ)p(σ0 )g (σ, σ0 )(σ + σ0 )2 dσdσ0 . (5.10)

8

Comparing Eqs. (5.9) and (5.10) leads to the simple relationship

Y = 2 (Z − 1) . (5.11)

This is the two-dimensional version of a result derived by Smith and Labík in Ref. [126].

Inserting the SPT result for βµex (λσ) [Eq. (5.7)] in to Eq. (5.8) yields

2ηs

Y = + 2Z η (5.12)

(1 − η)

where the packing fraction η = πρ〈σ2 〉/4 and s = 〈σ〉2 /〈σ2 〉. Equating Eqs. (5.11) and

(5.12) furnishes the final expression for the compressibility factor:

1 + (s − 1)η

Z= . (5.13)

(1 − η)2

π

Z = 1+ (1 + s)〈σ2 〉ρ + . . . (5.14)

4

which yields the correct second-virial coefficient (equal to half of the average ex-

cluded volume)

1 π π

Z Z

B2 = p(σ)p(σ0 ) (σ + σ0 )2 dσdσ0 = (1 + s)〈σ2 〉. (5.15)

2 4 4

η)2 = 1 + 3η/2 + . . .. Note that for the monodisperse hard-disk fluid, s = 1 and the

familiar SPT result Z = 1/(1 − η)2 is recovered.

We derive a simple, ME estimate of p k (σ), this being the joint probability distribution

of neighbors for a disk with given diameter σ. The overall degree distribution p k

is then obtained by integration of p k (σ). The derivation relies on there being con-

straints on the mean and mean-square number of neighbors for a disk with given

diameter. A very small disk is essentially an ideal point particle, and because such

small disks are so numerous, the neighbor network will be essentially random with

moments that will not depend sensitively on σ. (The degree distribution for Voronoï

tesselations of random sets of points in a plane – the 2D Poisson-Voronoï tesselation

5.2 Model and methods 80

– is known [129, 130].) On the other hand, a very large disk is expected to have an

average number of neighbors which scales linearly with σ. We make the following

ansatz; the average number of neighbors for a disk with diameter σ is

P

k kp k (σ)

K 1 (σ) ≡ = a 1 + b 1 f (σ) (5.16)

p(σ)

P

where we have used the fact that k p k (σ) = p(σ), and the sums are restricted to k ≥ 3

because triangles are the smallest polygons in the tesselation. a 1 and b 1 are parame-

ters, and f (σ) is a function with the properties f (0) = 0, and limσ→∞ [ f (σ)σ0 /σ] = 1.

Given the form of p(σ), an obvious choice for f (σ) is

(σ/σ0 )α+1

f (σ) = . (5.17)

1 + (σ/σ0 )α

P 2

The mean-square number of neighbors, K 2 (σ) = kk p k (σ)/p(σ) is assumed to be

such that the width of the distribution for a given σ is constant:

K 2 (σ) − K 12 (σ)

= a2 . (5.18)

K 12 (σ)

Z

p(σ)K 1 (σ)dσ = 〈k〉 (5.19)

Z

p(σ)K 2 (σ)dσ = 〈k 2 〉 (5.20)

where 〈k〉 = 6 (Euler) and 〈k 2 〉 are the averages for the whole network, and 〈k 2 〉 is to

be taken from simulation. This leads to the relationships

〈k〉 − a 1

b1 = (5.21)

〈 f (σ)〉

〈k 2 〉

a2 = −1 (5.22)

a 12 + 2a 1 b 1 〈 f (σ)〉 + b 12 〈 f 2 (σ)〉

be less than 6 because the average of K 1 (σ) over the whole diameter distribution –

including the large disks – should be precisely 6. For the present diameter distribution

p

(5.1) with α = 4, 〈 f (σ)〉 = 1/(2 2) and 〈 f 2 (σ)〉 = 21/32. We now seek an ME solution

5.3 Results 81

X

p k (σ) = p(σ); (5.23)

k

X

kp k (σ) = p(σ)K 1 (σ); (5.24)

k

k 2 p k (σ) = p(σ)K 2 (σ).

X

(5.25)

k

P

Maximizing the entropy S = − k≥3 p k (σ) ln p k (σ) with respect to p k (σ) (with the

method of Lagrange multipliers) leads to the result

£ ¤

(5.26)

where the Λi (σ)s are adjusted to satisfy the constraints (5.23-5.25). Finally, p k is

obtained from the relation Z

pk = p k (σ)dσ. (5.27)

5.3 Results

Reliable simulation results were obtained for pressures up to P ∗ = 2, corresponding

to a packing fraction η ' 0.45. For higher pressures the cluster moves were seen to

result in point reflections of increasingly large clusters, which ultimately precludes

effective equilibration of the system. Presumably there is no crystallization at very

high pressure. The equation of state – plotted as P ∗ versus η – for the polydisperse

hard-disk fluid is shown in Fig. 5.4. The raw data is given in Table 5.1. The SPT (5.13)

and virial-expansion (5.14) results are shown for comparison. The SPT is surprisingly

good, being accurate for packing fractions up to about 0.3.

Simulation snapshots at P ∗ = 1 are shown in Fig. 5.5; the packing fraction at this

pressure is η ' 0.36. For clarity, only a subdomain (containing 1600 particles) of a

particular configuration is shown. Three different views are displayed: (a) the actual

hard-disk configuration; (b) the radical tessellation; and (c) the Delaunay triangles

formed by the links between neighboring vertices (defined by disks that share cell

edges).

On the basis of the radical tessellations, the degree distribution p k was con-

structed, being the probability that a disk has k neighboring disks. Given that ten

5.3 Results 82

Figure 5.4: Equation of state – reduced pressure P ∗ = βP σ20 versus packing fraction

η: MC simulations (points); SPT (solid line); virial expansion with B 2 (dashed line).

configurations were simulated, each containing 104 particles, the lower limit on p k

accessible from our simulations is ∼ 10−5 . In all cases it was confirmed that the aver-

age number of links, 〈k〉, is precisely equal to six. Figure 5.6 shows p k as measured

and averaged from simulations at P ∗ = 0.01, 0.1, 0.5, 1, 1.2, and 2. Error bars on p k

represent the statistical uncertainty in the mean obtained from averaging over the

ten distinct configurations. The small-k portions of p k (k & 10) show only small

variations over the entire range of pressures simulated. At low pressures (P ∗ < 0.5) p k

shows no sign of a long tail. At such low packing fractions, the particle positions are

uncorrelated and the degree distribution resembles that of the 2D Poisson-Voronoï

tesselation [129, 130]. At pressures in the range 0.5 ≤ P ∗ ≤ 1 it was observed that the

tail of p k for k & 21 appears as a straight line on a log-log plot, indicating consistency

with the power-law decay k −γ . At higher pressures, no power-law tail is apparent over

the range of k accessible in simulations. For all pressures P ∗ ≥ 0.1, a kink is apparent

in p k at the point where p k ' 10−4 . Since the proportions of particles contributing to

p k before and after the kink appear constant, this suggests that p k is partitioned in to

‘small-particle’ and ‘large-particle’ contributions, the latter possibly corresponding

to an asymptotic scale-free regime.

Figure 5.7 shows the tails of p k as measured in simulations with pressures in the

range 0.5 ≤ P ∗ ≤ 1. The kinks at p k ' 10−4 are more clearly visible. The asymptotic

decay of p k is consistent with a power-law, but it is far from being unambiguous. It is

difficult to access larger values of k because of limitations on the number of particles

that can be simulated; this restricts the form of the diameter distribution, and in

5.3 Results

Figure 5.5: (Color online) Sub-domain (N = 1600 particles) of a simulation configuration at P ∗ = 1 and η ' 0.36: (a) the actual hard-disk

configuration; (b) the radical tessellation; and (c) the Delaunay triangles formed by the links between neighboring vertices.

83

5.3 Results 84

pressures in the range 0.01 ≤ P ∗ ≤ 2.

least-squares fitting of power laws to the tails of p k yields the lines shown in Fig. 5.7.

The corresponding fitting ranges and exponents (γ) are reported in Table 5.1.

It has been pointed out that the least-squares fitting procedure may give rise to

biased, and hence erroneous, results for the power-law exponent (γ) due to large

fluctuations in the tail of the distribution [131, 132]. There is an alternative fitting

scheme based on a maximum likelihood estimator (MLE), which for power-laws

yields a simple, closed-form expression for the exponent [131, 132]. We implemented

this scheme on the cumulative probability, c k = ∞ p k ∼ k −γ+1 , which serves to

P

k

reduce statistical errors. We report the resulting MLE exponents in Table 5.1; the

values are slightly smaller than, but generally consistent with, those obtained from

least-squares fitting.

Interestingly, γ is relatively insensitive to the packing fraction, being in the region

5-6; the uncertainties quoted in Table 5.1 are associated with the fitting procedure,

and so they are underestimates. Naïvely, if the number of neighbors is proportional

to the disk diameter, then one might expect p k ∼ k −4 ; in reality, γ is significantly

greater than 4. The apparent exponent γ > α can be understood by comparing the

MC simulation results with the predictions from the ME theory derived in Section

5.2.3. Figure 5.8 shows p k from simulations with P ∗ = 0.5 and P ∗ = 1, along with

plots of Eq. (5.27) computed using the best-fit parameters a 1 = 5.50 and a 1 = 5.39,

respectively, and values of 〈k 2 〉 reported in Table 5.1. The ME results are in quite

good agreement with those from simulations, although the kink at p k ∼ 10−4 is not

5.3 Results

0.01 0.0078(2) 6.00 37.758(4) 0.2210 0.475(8)

0.1 0.068(1) 6.00 37.78(7) 0.2220 0.472(14)

0.3 0.171(2) 6.00 37.88(3) 0.2285 0.467(10)

0.5 0.245(3) 6.00 38.18(3) 0.2445 6.0(1) 5.7 21 ≤ k ≤ 28 0.456(9)

0.6 0.276(3) 6.00 38.33(4) 0.2525 5.8(1) 5.7 22 ≤ k ≤ 29 0.448(7)

0.7 0.302(3) 6.00 38.44(4) 0.2581 5.9(1) 5.6 23 ≤ k ≤ 31 0.441(7)

0.8 0.324(3) 6.00 38.55(5) 0.2641 5.7(1) 5.5 23 ≤ k ≤ 32 0.433(7)

0.9 0.346(3) 6.00 38.67(4) 0.2698 5.6(1) 5.4 24 ≤ k ≤ 33 0.424(7)

1 0.362(2) 6.00 38.76(4) 0.2744 5.6(1) 5.4 24 ≤ k ≤ 34 0.420(6)

1.2 0.389(2) 6.00 38.67(7) 0.2724 0.408(3)

2 0.454(6) 6.00 39.0(1) 0.2901 0.377(6)

p fluid with α = 4: packing fraction η; mean and mean-square degree,

2 2

〈k〉 and 〈k 〉; relative width of the degree distribution, w = 〈k 〉/〈k〉2 − 1; power-law exponent, γ, for the tail of p k (and the range over

which power-law behavior is observed) from least-squares fitting (LS) and a maximum likelihood estimator (MLE) [131, 132]; assortativity r .

Figures in brackets are statistical uncertainties in the last digits based on one standard deviation.

85

5.3 Results 86

reduced pressures in the range 0.5 ≤ P ∗ ≤ 1. The lines are fitted power-law extrapola-

tions p k ∼ k −γ ; the least-squares exponents are reported in Table 5.1.

captured so well. But the main points are that in the region of k = 20-30, where

the simulation results are consistent with a power-law decay, the ME predictions

have apparent exponents greater than 4 (at k = 20, γ = −d ln p k /d ln k ' 5.8), and

that as k → ∞, p k approaches the anticipated k −4 behavior (at k = 100, γ ' 4.3). Of

course, this is largely a consequence of the prescription for K 1 (σ). But the agreement

between simulation and theory at moderate k suggests that, at large enough k, the

simulated network degree-distribution would follow the power law p k ∼ k −4 .

5.3.3 Assortativity

of vertices with equal degrees to cluster. A convenient measure of assortativity is

clearly defined in Ref. [101], but for clarity the derivation is reproduced here. The

degree distribution of vertices attached to randomly selected edges is not equal

to p k , because high-degree vertices have more links. Instead, the distribution is

proportional to kp k , and the distribution of remaining vertices (not including that

attached to the randomly selected edge) is q k ∝ (k + 1)p k+1 . The assortativity can

then be related to the joint probability distribution function of there being j and k

remaining vertices at each end of the selected edge [100]. Denoting this function

by e j k , the following properties are to be noted: for an undirected graph e j k =

P P

e k j ; j k e j k = 1; and j e j k = q k . For networks which are neither assortative nor

5.3 Results 87

(solid points and solid line) and P ∗ = 1 (open points and dotted line) from MC

simulation (points) and the ME theory (lines).

disassortative, the degrees of the vertices at the ends of the edge are uncorrelated

and so e j k = q j q k . A correlation function between degrees can therefore be defined

P

as 〈 j k〉 − 〈 j 〉〈k〉 = j k j k(e j k − q j q k ) where the angled brackets denote an average

over all edges in the graph. For the comparison of diverse graphs, it is convenient

to normalize this correlation function by its maximal value, which corresponds to

the case of a perfectly assortative network, i.e., the degrees of connected vertices are

identical. In this case e j k = q k δ j k and j k j k(e j k − q j q k ) is equal to the variance, σ2q ,

P

defined as P

jk j k(e j k − q j q k )

r= (5.28)

σ2q

where r ranges from −1 for a perfectly disassortative network, through 0 for a network

with no disassortative or assortative mixing (such as a random network [85, 101])

to +1 for a perfectly assortative network. The average values of r obtained for the

hard-disk neighbor networks are reported in Table 5.1. The results indicate that the

hard-disk fluid is strongly assortative at all pressures considered. If the degree of

a disk is dictated by its diameter, then in physical terms, the results indicate that

‘small’ disks are preferentially solvated by other small ‘disks’; the ‘large’ disks – with

high degree – are more likely to be solvated by ‘small’ disks, due to the form of the

diameter distribution. Nonetheless, some weak clustering of large disks is apparent

in Fig. 5.5, particularly in 5.5(b) and 5.5(c); note the associations between different

5.4 Conclusions 88

‘sparse’ regions of the network. The physical driving force for the clustering of large

disks might be identified with a depletion interaction [133, 134, 135] mediated by

the osmotic pressure of the small disks, but of course what constitutes the boundary

between ‘large’ and ‘small’ is not well defined. r decreases slowly and monotonically

with increasing packing fraction. One possibility is that as the packing fraction is

increased, particle packing and correlations become more pronounced, and the

‘large’ disks develop links with an increasing number of neighboring ‘small’ disks,

and the assortativity decreases. This process is, of course, accompanied by a broad-

ening of the degree distribution, as illustrated in Fig. 5.6, and as quantified by the

p

relative width, w = 〈k 2 〉/〈k〉2 − 1, reported in Table 5.1. The broadening of p k , and

the resulting effects on q k , are already taken in to account in the definition of r

[Eq. (5.28)].

The assortativities of real networks vary widely. Most ‘social’ networks, such as

scientific co-authorships and movie-actor collaborations, are clearly assortative, with

r values up to about 0.4 [101]; successful individuals are often attracted to others. On

the other hand, the networks represented by the World Wide Web, the Internet, and

most significantly, many biological situations are disassortative (with r values down

to about −0.3) [101]; this property may make the network more resilient to random

attack, since the hubs (which may play crucial roles in the function of the network)

are not connected [92]. The present simulation results indicate that if scale-free

neighbor networks do exist in the polydisperse hard-disk fluid at pressures in the

range 0.5 ≤ P ∗ ≤ 1, then they are strongly assortative, with values of r comparable

to those found in social networks. It is tempting to speculate that the analogues of

groups and communities in social networks [103] are ‘clusters’ of small and large

disks in the polydisperse fluid.

5.4 Conclusions

simulations of a polydisperse fluid of hard disks. Each disk represents a vertex

on a graph, and the edges of the graph correspond to links between neighboring

disks as identified by a radical tessellation. Working on the naïve assumption that

the number of links is proportional to the disk diameter, an asymptotic power-

law diameter distribution was expected to give an asymptotic power-law degree

distribution. The simulation results are consistent with power-law decay – within

the limits imposed by the simulated diameter distribution – but only for a specific

range of disk densities in which particle correlations are sufficiently pronounced.

5.4 Conclusions 89

The apparent exponent of the degree distribution γ was found to depend weakly on

the disk density, and to differ from the exponent of the diameter distribution α. A

comparison with predictions from a maximum-entropy theory suggest that this decay

is not the asymptotic one, and that for sufficiently large particles (not accessible in

simulations) γ would approach α. The results therefore suggest that, in principle,

the near-neighbor network in the polydisperse hard disk fluid may be scale-free. It is

anticipated that a similar situation would obtain with different choices of α, but of

course α should be as small as possible if a sufficient number of hubs are to be found

in a network of finite size. Earlier results [112] and the form of the maximum-entropy

theory suggest that non-power-law diameter distributions will not give rise to a scale-

free neighbor network. The assortativity of the network was found to be positive over

the whole range of disk densities, reflecting a preference for links between vertices

of equal degree. In other words, ‘small’ particles are preferentially solvated by other

‘small’ particles, and ‘large’ particles show some tendency to cluster together. Finally,

some effort was directed towards analyzing the thermodynamic properties of the

fluid. No evidence for crystallization at high packing fractions was found. A simple

equation of state – derived from a scaled-particle theory – was demonstrated to be

accurate up to moderate packing fractions.

In stochastic models of networks, the position of a vertex on a graph does not

influence its degree (i.e., the number of other vertices with which it shares edges).

But in the case of hard disks at equilibrium, the degree of a disk is dictated by its

size and the sizes of its immediate neighbors, through packing constraints. The

spatial correlations in fluids, and therefore the properties of the neighbor network,

are dictated by the requirement to minimize the free energy of the system. Hence,

the apparent scale-free properties of the network are static, and do not arise from

any dynamic mechanism, such as preferential attachment. It would be interesting to

find out whether the formation of assortative, scale-free networks in other contexts

can be rationalized using equilibrium statistical mechanics.

An experimental realization of a polydisperse ‘fluid’ could actually be a gran-

ular material, with a scale-free size distribution achieved by milling and grinding.

Alternatively, by analogy with cellular networks, froths and foams produced with

an element of bubble division and disappearance may exhibit a scale-free degree

distribution [98, 99]. It should be pointed out, however, that in conventional colloidal

fluids, polydispersity is usually better described by log-normal or Γ distributions.

The physical properties of two-dimensional materials possessing scale-free networks

may be of interest, and demand further study.

Chapter 6

mixtures of dipolar colloids in tilted

fields

6.1 Introduction

has recently attracted a great deal of attention [136, 137, 138, 139, 140, 141, 142,

143, 144] due to the large number of potential applications ranging from food and

agrochemicals to petrochemicals and pharmaceuticals. Unlike particles in the bulk,

the effective interactions between colloidal particles at fluid interfaces are influenced

not only by the properties of the particles and the solvent, but also by the surface and

line tensions of the interface [143].

From various experimental realisations of binary mixtures of dipolar colloids,

two such realisations are related with this work and we briefly describe them here.

The first realisation deals with colloidal dipoles, which are roughly spherical particles

with diameters in the range 10 nm - 1 µm. For preventing irreversible aggregation

the particles are usually coated with a thin layer of non-magnetic material. Thus

the net colloid-colloid interaction is mainly dipolar, with the coatings providing a

relatively weak short-range dispersion interactions. The colloidal dipoles are sus-

pended in a pendant drop in gravity (system I), see Fig. 6.1(a). In the second, colloids

of polystyrene colloids float on an oil-water interface (system II), see Fig. 6.1(b) .

The colloids are trapped at the oil-water interface because this way they minimise

the surface between the two phases. Nevertheless, the polystyrene colloids do not

have a molecularly smooth surface and water molecules are trapped at the colloid-

oil interface. This has as a consequence the formation of water droplets or a thin

90

6.1 Introduction 91

Figure 6.1: Scematic illustration of system I and II; (a) two colloids of different size

trapped at the oil-water interface. The dipole moments P 1 and P 2 are given by the

vector sum of the dipole moments on the particle-oil interface; (b) colloidal particles

trapped at the water-air interfacein a pendant water droplet. The external magnetic

field B is applied on the colloids.

film surrounding the polystyrene colloid with dipoles. The net dipole moment of

the dipoles surrounding the colloid is non-zero, because the dipole interaction is

screened in the aqueous phase.

In both cases, one can apply an external magnetic field to induce a repulsive

colloid-colloid interaction, whose direction and strength can easily be controlled via

the external field; the dipole moments almost perfectly align along the field direction

and the magnitude of the dipole moment is proportional to the field strength. In the

conventional setup, the field is perpendicular to the interface, inducing a repulsive

interaction between the colloids, additional to the short-ranged steric repulsion

arising from the physical cores of the colloids. For large fields, the typical strength

of the repulsive dipole-dipole interaction is much larger than the thermal energy,

leading to crystallisation of the colloids. Both realisations have been extensively

studied with experiments [136, 138, 140, 139], and simulations [141, 137, 144]. As

shown in [142] at suffiecient low density the dipole-dipole interactions of system I

and II become equivalent. The pair interactions of colloids of a binary mixture (with

diameter σA or σB ), has the form of a power law

Ψi j (r ) ∼

=(z i z j )5/2 /r 3 (6.1)

where r is the distance between the centers of the two colloids, and z i , j is either 1 or

z = σB /σA < 1 if i , j = A or B. At this regime where system I and II become equivalent

the z 5/2 corresponds to the suspensibility of the supermagnetic colloids of system I.

In this work we will work in this density regime where eq. 6.1 holds.

Knowing the effective interactions between the particles one can use computer

simulations to obtain the minimum energy configuration (MEC), and thus the fa-

6.2 Simulation Methods 92

vored crystal structure at T = 0. However, it is well known that binary mixtures tend

to exhibit a broad spectrum of rather complex alloy phases [145] corresponding to

a very rough energy landscape; few optimization algorithms are able to solve such

high-dimensional and complex problems. To overcome this difficulty, genetic algo-

rithms (GAs) have been applied to the case of binary mixtures with their interactions

been described by to eq. 6.1. The results revealed a rich spectrum of exotic ordered

ground-state structures [142].

The main focus of the current work was to explore, with the use of GA, the crystal

structures emerging from a binary mixture of colloids when the external field is tilted

with respect to the normal of the interface. The tilt angle introduces an anisotropic

interaction creating new exotic crystal structures. As the field is tilted beyond a

critical angle the patterns switch over being dominated by chains aligned parallel to

the field.

The paper is organized as follows. In Sec. 6.2 we define our model and present

the simulation methods used. The results are presented and discussed in Sec. 6.3.

Finally, in Sec. 6.4 we summarise the conclusions.

6.2.1 Model

We consider a binary mixture of particles, namely type A and B, with their diameter

ratio, z = σB /σA = 0.5, unless stated otherwise. To determine the global energy

minimum of such a system, we assume that the particles arrange themselves in a

set of Bravais lattices. Specifically, we consider a two-dimensional Bravais lattice

with an N -particle unit cell, where N is the sum of particles of type A and B, that is

N = NA + NB .

The candidate structures are generated by the periodic repetition of the unit

cell, which is a parallelogram formed by the vectors a and b that substain an angle ψ

between them, as shown in Fig. 6.2(c). For convenience, but without loss of generality,

we set the particles occupying vertices of the parallelogram to be of type A, and their

positions are described by the linear combination r = l a + mb, where l and m are

integers. We introduce a coordinate system (x, y) and, choosing the x axis parallel to

the unit-cell vector a, the primitive vectors of the Bravais lattice take the form:

a = α (x, 0)

b = α cos ψ, sin ψ

¡ ¢

(6.2)

6.2 Simulation Methods 93

Figure 6.2: (a) and (b) show schematic views of two particles, with magnetic moments

m and m0 , perfectly aligned along the external magnetic field, B. (c) The geometry

of a two-dimensional Bravais lattice with one particle per unit cell, as well the basis

vector ck that characterizes the position of an additional particle.

where x is a variable determining the length of a vector, |a| = x|b|, and α is a scal-

ing variable defined by the number density as ρ = N /|a × b| = N /xα2 sin ψ. Now,

since ρ will be fixed, it allows the elimination of one parameter of our model,

α = N /xρ sin ψ. We will call the lattice just described as the primary lattice. The

p

additional particles of the unit-cell, which can be of type A or B, are placed inside the

parallelogram, and their positions are specified by the set of vectors ck = c ka a + c kb b,

where c ka,b are coefficients and k is the index of the additional particles, k = 2, ..., N .

The value k = 1 is reserved for the coordinates of the particle belonging to the first-

lattice, c1 = (0, 0).

When such a cell is repeated periodically over space, it produces N lattices, and

each of them is shifted with respect to the first-lattice by the vectors ck . Hence, the

positions of the particles on the (x, y) plane are given by the vector r + ck .

To calculate the total energy per particle, U , in a structure defined by our model,

one has to calculate the lattice sum of all unique interactions between the particles

of the unit cell; the interactions are defined in Sec. 6.2.2, but for now they will

represented by u i , j (r). For example, if we had one particle in the unit cell, in other

words we have only the primitive lattice, then the total energy per particle would be:

1X

Ui1,i = u i i (r) (6.3)

2 r6=0

where the set {r} spans a Bravais lattice, which takes the form of a lattice sum

P∞ P∞

l =−∞ m=−∞

(...) running over the integers l and m, with the exception of (l , m) =

(0, 0). Having an additional particle in the unit cell, the total energy per particle will

be the average of the sum of each particle of the unit cell interacting with all other

particles within its own sublattice, plus the additional energy between the particles

belonging to different sublattices. Therefore the total energy per particle for two

6.2 Simulation Methods 94

1³ 1 ´ 1X

U2 = Ui ,i +U j1, j + u i , j (r + c2 − c1 ) (6.4)

2 2 r

Now, with the same reasoning we write the total energy per particle for an N -particle

unit cell:

"

1 1X X

UN = u i ,i (r)

N 2 r6=0 i =A,B

#

X NX N

−1 X

+ u i , j (r + ck − ck 0 ) (6.5)

r k=1 k 0 =k+1

Hence, the parameters that characterise the structure of the lattice are N , ρ, x, ψ,

c ka , and c kb . These parameters are subject to the following constrains:

Note that the representation in eq. 6.2, even with the above constraints, is not unique,

since different sets of parameters can create equivalent structures; these ambiguities

are excluded by a suitable algorithm described in Sec. 6.2.3.

In the following we characterize composition of the binary mixture by the mole

fraction of small particles, C = NB /N . The parameter ranges we explore in this work

are, ρ = 0.4, 0.6 and 0.8, C = 1/2, 2/3, 3/4, 4/5, 1/3, 1/4, and 1/5.

6.2.2 Interactions

tions, and with an external magnetic field, B, strong enough to create crystallization.

Before we define the interactions between the particles, we define the two compo-

nents of the B, one parallel to the plane of the interface, B∥ , and the other normal to

the interface, B⊥ , see Fig. 6.2(a). For convenience, we set the B∥ to be parallel with

the x-axis and the vector a. Therefore the field is expressed as B = |B|(cos φ, 0, sin φ).

Now, based on the approach presented in Ref. [146], the dipole-dipole pair potential

at the interface in the presence of a strong magnetic field, B, is

χi χ j B2 1

Vdd (r, χi B, χ j B) = (1 − 3 cos2 φ cos2 θ) (6.7)

2 r3

on the lattice, φ is the tilt angle of the magnetic field with respect to the confining

6.2 Simulation Methods 95

Figure 6.3: (Color online) Two diagrams showing the total potential of interaction at

θ = 0 between two particles of type A (top) and one A and one B type (bottom) as a

function of interparticle separation for different values of the tilt angle, φ.

However, since we will work in low enough density, where the asymptotic behav-

ior for a system of dipolar colloids when the external magnetic field is perpendicular

to the interface holds and is described by eq. 6.1, then we recall the correspondence

χi χ j B2

z i5/2 ↔ χi . Using this, we eliminate the factor 2

in eq. 6.7, and thus we can

re-write the pair dipole-dipole interaction as:

1

Vdd (r, z i , z j ) = (z i z j )5/2 3

(1 − 3 cos2 φ cos2 θ) (6.8)

r

The degree of anisotropy is controlled via the tilt angle φ, as seen in Fig. 6.4, which

allows the particles to adopt less repulsive head-to-tail configuration (θ = 0). In this

p

work we will restrict the tilt angle above the threshold value arccos(1/ 3) ≈ 54.7◦ ;

below that value, the dipole interactions in the nose-to-tail configuration becomes

attractive. The values we explore are φ = 54.7◦ , 59◦ , 64◦ , 74◦ , and 84◦ . Additionally,

we generate structures at 90◦ for comparison.

At high degree of anisotropy the particles will collapse to each other due to

the absence of repulsive interactions in the x direction. For this reason the purely

repulsive Weeks-Chandler-Andersen (WCA) potential [59], which is a Lennard-Jones

potential cut and shifted at the position of the minimum, r min = 21/6 σi j , is used to

6.2 Simulation Methods 96

Figure 6.4: (Color online) The diagram shows the total potential of interaction

between two A type particles at fixed interparticle separation, r = 1.2σ A , as a function

of θ.

model the short-ranged repulsion between the physical cores of the colloids,

·³ ´ ³ σ ´6 ¸

σi j 12

+ εi j

ij

4ε

ij r − r r ≤ r min

VWCA (r ) = (6.9)

0 r > r min

where σi j is the effective particle diameter, and εi j sets the energy scale of the short-

ranged soft core repulsion; the values of these parameters were determined by the

p

well-known Lorentz-Berthelot mixing rules [147], ²i j = ²i i ² j j , σi j = (σi i + σ j j )/2.

Joining eq. 6.8 and eq. 6.9 together, the total pair potential takes the form:

Introduction

hard optimization problem, we need to search through a space of potential solutions

by either finding, if possible, an efficient problem-specific algorithm which is approx-

imately optimal, or by using probabilistic algorithms, which despite being powerful,

do not guarantee to give the optimum solution. For small spaces, classical exhaustive

methods usually suffice; for larger spaces special artificial intelligence techniques

6.2 Simulation Methods 97

must be employed. GAs are stochastic search algorithms invented in the 1970s by

Holland to solve high dimensional and complex problems in engineering science

[148]. These methods are inspired by the Darwinian struggle for survival and employ

evolutionary processes as key elements to obtain the optimal solution to a particular

problem. However, strictly speaking, GAs do not optimize, and neither does biologi-

cal evolution [148, 149]. Evolution uses whatever material that is at its disposal to

produce above average individuals; it has no ultimate goal of “perfection”. Neverthe-

less, GAs form the basis of extremely robust optimization methods, which are able to

take the whole search space into account, and at the same time concentrate their

computing efforts on promising regions (for an overview see Ref. [150]).

We briefly summarize the basic ideas of a GA and in the subsequent section will show

how it will be applied to our specific problem.

GA represents a candidate solution as an individual, whose genes, fixed in number,

represent the variables of the given problem. The genes in an individual can take

values from a suitably chosen alphabet, such as binary, with digits 0 and 1 as its

letters. All the genes together form the chromosome as a string of digits. Individuals

are evaluated via a problem-specific fitness function in the sense that a better or

fitter individual has a higher fitness value. A large, but fixed, number of individuals

form a generation.

The initialization of the GA’s first generation is realized at random. A new gen-

eration is formed by selecting individuals to be parents from the former generation

according to their fitness values, and then generating new individuals to form the

subsequent generation. Several procedures for this mating recombination processes

have been proposed in the literature [151]. In addition, mutations are performed with

a certain probability during the mating process, a procedure necessary to avoid per-

sistent inbreeding and to allow at the same time for the reintroduction of new or lost

genetic material. At each iteration selection-recombination-mutation-evaluation

cycle, a fraction of the former generation is replaced to form the new generation.

Usually, we retain individuals with the highest fitness value, but other replacement

procedures exist. The final solution is represented by the fittest individual in the

population after many generations.

These general ideas behind GAs are now applied to our problem of obtaining the

optimal crystal structure of a binary mixture of colloidal particles, by using the model

6.2 Simulation Methods 98

described in sec. 6.2.1. The formalism has been used in similar problems elsewhere

in the literature [152].

We translate our model represented by the set of parameters {x, Ψ, c 1a , c 1b , ..., c N

a b

, cN },

where Ψ = ψ/(π/2), into an individual I = I (y 1 , y 2 , ..., y 2N ), where y 1 , .., y 2N repre-

sent the parameters of our model as strings of genes of fixed length, n gene = 6. The

genes are expressed by a decimal alphabet i.e. an alphabet composed by the one-digit

integers 0, 1, 2...9. The aim of the encoding is to produce a chromosome from which

the GA will use later. For example, assuming an individual is a point in (x, y) space;

the encoding process would produce

For the present investigation, we used the number of individuals, N p = 500, and the

number of generations NG = 300. Even though for most cases explored a smaller

genetic population would be suffient to give a solution; a large population maintains

diversity between individuals, and thus covering the whole search space.

As mentioned in Sec. 6.2.1, we need to implement an algorithm to ensure a unique

representation of the Bravais lattice. This is achieved by requiring the perimeter of

the unit cell, Π, to be minimal, and it is given by the equation,

To minimize eq. 6.12, the following iterative algorithm was applied: the vectors a and

b are replaced by new vectors, from the four options the combination with the lowest

value out of Π,

{a ± b, b} and {a, a ± b} (6.13)

If none of the possible combinations minimize Π, then the algorithm stops and

the initial a and b are used. The vectors ck are not required to be checked by the

algorithm, since they are linear combination of a and b.

A positive fitness value is assigned to each individual I via the fitness function

f (I ): a higher fitness value characterizes a better solution. The optimum solution is

a crystal structure with the lowest free energy, F = U − T S, where S is entropy, but for

our purposes we look for crystals formed by the system at zero temperature (T = 0).

Therefore, the optimum crystal structure will have lowest potential energy, U , which

can be calculated by eq. 6.5 and eq. 6.10. Additionally, for practical reasons we look

at the ratio U (I )/U (I hex ), where U (I hex ) is the potential energy of a hexagonal

structure (hex) generated by one particle unit cell at the same density. The reason

behind this is to ensure that the values used will be of order 1. Hence, f (I ), in our

6.2 Simulation Methods 99

problem, reads

U (I )

½ ¾

f (I ) = exp 1 − (6.14)

U (I hex )

Once the first generation, G0 is realized and its individuals are evaluated by the

fitting function, we sort them out by a ranking scheme. For example, the fittest

£ ¤ £ ¤

individual will have rank value R f (I j ) = 1, while the least fit will have R f (I j ) =

N p . This ranking scheme is important for avoiding breeding shortcomings and to

ensure the existence of selection pressure throughout the simulation [151].

Afterwards, the interactive process of creating a new generation, starts. In the

first step, two individuals are selected as parents according to the rank of their fitness

values with the roulette wheel algorithm [151]. First, we generate a random number

w ∈ [0, S], where S is the sum of the differential fitness, S j , over all the individuals in

the population:

Np Np

£ ¤

X X N p − R f (I j ) + 1

S= Sj = (6.15)

j =1 j =1 Np

T j −1 ≤ w < T j (6.16)

j

X

Tj = S j , j = 1, ..., N p (6.17)

i =1

1

is occupying a sector of angular size S /S

2π j

and the random number represents the

spinning wheel. In other words, the probability of the wheel stopping at a sector

is proportional to the sector. Once the parents are chosen, two new individuals

populate the new generation via a one-point crossover, as visualized in Fig. 6.5.

The crossover point, which is randomly chosen, is where each of the parent’s gene

sequence is cut and combined to create the new individuals. This action is unbiased

in respect to the geometry of the problem and cannot be associated with any physical

movement of the particles. Further, we perform mutations with a probability p m ,

which has a typical value of 0.1, on the new individuals. It means that we randomly

choose one of the genes and replace it with a new one.

Even though the process of recombination and mutation will create two new

individuals, it is not guaranteed that they will join the new generation. A number of

criteria must be fulfilled. First, they must be at least as fit as the least-fit members

of the existing population; the latter will be eliminated and replaced by the new

6.3 Results & Discussion 100

Figure 6.5: (Color online) Schematic representation of the one-point crossover that

generates two new individuals from two parents.

members. Second, the new individuals must not be identical with any other member

of the population; this no-duplicate criterion is an important safeguard against

inbreeding.

To test the efficiency of the reproduction strategy described above, it was com-

pared to an alternative reproduction strategy in which we select parents randomly

with no regard to fitness, and then randomly replace members of the population with

the new individuals. An exception to this replacement is granded to the fittest indi-

vidual, as a form of elitism. Similar strategies to this has been found advantageous

for problems tending to suffer from premature convergence [153].

After the GA has provided with a solution a final optimization with gradient

descent was applied. The last step was needed to increase the accuracy of the results.

We present the ordered equilibrium configurations obtained by the GA, as the angle

of the magnetic field is tilted. The results are organizing into four blocks. (a) In

Fig. 6.6 structures with C ≥ 1/2 are shown. (b) Structures with C < 1/2 are in Fig. 6.8.

(c) A comparison with structures of different densities are in Fig. 6.9. (d) Structures

with smaller diameter ratio z = 0.3296, at ρ = 0.6 are presented in Fig. 6.10.

6.3.1 C ≥ 1/2

We start with the case C ≥ 1/2, where the small particles form more complex struc-

tures, as seen in Fig. 6.6. For the angle range 74◦ < φ < 90◦ , the patterns of the

structures only slightly perturbed from the structures at φ = 90◦ . For φ < 74◦ the

anisotropy becomes significant, and both the small and big particles attempt to take

advantage of it by generating new patterns; see Fig. 6.6(i), (m), (n), (r) and (s). At

90◦ 74◦ 64◦ 59◦ 54.7◦

C = 1/2

6.3 Results & Discussion

C = 2/3

C = 3/4

C = 4/5

Figure 6.6: (Color online) Minimum energy configurations for C ≥ 1/2, for ρ = 0.6.

101

6.3 Results & Discussion 102

Figure 6.7: (Color online) The diagram shows relative change of length of the vector

a = αx for different tilt angles.

φ = 54.7◦ there is no repulsion along the x-axis and the particles form chains parallel

to the field.

To understand the results, we view the length of vector |a| as a function of ∆φ =

90 − φ, so at ∆φ = 0◦ , |a| = |a|0 . Naively, one would expect the length of vector a to

◦

Fig. 6.7, we plot the ratio of relative displacement of the A-type particles of the first

lattice along the x-axis, (|a|0 −|a|∆φ /|a|0 ), versus ∆φ. For all cases, the ratio remains at

the same level up to the threshold ∆φ ≈ 16◦ , above that threshold it rapidly increases.

However, in the case C = 3/4, initially there is an increase, at ∆φ there is descrease,

meaning that the length of the a vector increases, but after a certain value of ∆φ,

the ratio increases as expected. Note that in our initial assumptions of our model,

the length of vector a is always smaller than the length of b. Even though, this

significantly simplifies the problem it has also the effect to limit the search space.

Nevertheless, when the restriction was removed, there was no change in our results.

An explanation for this behavior is that the small particles in the unit cell feel first

the anisotropy and attempt to come in between the primary lattice’s particles into a

head-tail configuration, compare Figs. 6.6(l) and (m). Additionally, the particles of

the primary lattice also attempt to position themselves into a more favorable oblique

formation.

Now, as the angle of the magnetic field is tilted further the particles of the primary

lattice prefer to come closer together, while increasing their separation distance in

the y-axis, as seen clearly in the cases C = 1/2 and C = 2/3, Fig. 6.6(d) and (i). In

6.3 Results & Discussion 103

the cases C = 3/4 and C = 4/5 at 59◦ , we observe that one of the small particles has

moved between the particles of the primary lattice, by creating an inhomogeneous

chain, Fig. 6.6(n) and (s). The structures at φ = 54.7◦ converge into a series of chains

parallel to the magnetic field. The small particles can form chains of uneven numbers,

Fig. 6.6(o). The reason is that the primary lattice particles come as close as possible

with each other along the x-axis, thus maximizing their distance in y-axis, and by

doing that in essence they reduce the available space for the small particles to create

chains. As the number of small particles increases there is no enough space to

accompany the third particle, therefore it creates a new chain. For C = 4/5 two such

chains composed by small particles are formed. Even though chains can be formed

for φ > 54.7◦ , such in the cases Figs. 6.6(n) and (s), some of them have mixed type of

particles something that it is not observed at φ = 54.7◦ , see Figs. 6.6(e), (j), (o) and (t).

In the second block of results, C < 1/2, the big particles become the majority com-

ponent and attempt to form hexagonal structures, which are disturbed by the small

particle, as seen in Figs. 6.8(a), (f ) and (k). In particular, at perpendicular field,

φ = 90◦ , the neighbors of the small particle form a square, which is also seen as the

optimum structure for the system C = 1/2, see Fig. 6.6(a). Similar with the cases of

C ≥ 1/2, as the field is tilted the particles will rearrange by taking advantage of the

anisotropy.

When anisotropy becomes significant enough, the particles attempt to form

chains as seen in Figs.6.8(c), (d), (h), (i) and (m), (n). In the cases of C = 1/3 and

C = 1/5 the chains are straight, but in other case, C = 1/4, the particles’ excluded

volume comes into effect and restricts them to form straight lines. The result is to

form small zig-zags [frustrated chains], but this effect disappears as the repulsion

along the x-axis is reduced. Also, in the case of C = 1/5 see Figs.6.8(k-o), there

was no significant changes in the system’s structure due to the domination of the

big particles. The hexagonal structures change to obligue and then to chain-like

oblique structures, while the rectangular structures, created by the effects of the

small particles, do show any significant changes.

At φ = 54.7◦ the particles form chains of the same type, aligned parallel to the

field. It is interesting to note that parallel chains of particles of A-type form chain-like

oblique structures as it has been found for one component systems in Ref. [146], but

if between the chains there is a small particle, then the chains of A-type particles form

rectangular structures with be the small particle in the middle (compare Figs.6.8(e),

(j), and (o) with Fig.6.6(e)).

90◦ 74◦ 64◦ 59◦ 54.7◦

C = 1/3

6.3 Results & Discussion

C = 1/4

C = 1/5

Figure 6.8: (Color online) Minimum energy configurations for C < 1/2, for ρ = 0.6. The particles with color blue belong to the primary

lattice, while the particles with color red are A-type particles.

104

6.3 Results & Discussion 105

φ = 64◦ φ = 64◦ φ = 54.7◦ φ = 64◦

ρ = 0.8

ρ = 0.6

ρ = 0.4

as their patterns are affected by density.

potential scales as a power law at long distances, ∼ r −3 . This feature generates pat-

terns that are scale-free, meaning that the overall density is an irrelevant parameter,

as shown in previous investigations, where the field was perpendicular to the inter-

face [142]. Only at higher densities, does the excluded volume becomes a significant

factor, the system cease to be scale-free. In our simulations, we do observe the same

patterns for different values of densities for tilted external fields with non-zero in-

teraction along the x-axis. In Fig. 6.9 we present different compositions at different

densities with a tilted field, and discuss how it is affected by anisotropy. In the first

example, see Fig. 6.9(a), (f ), and (k), the system, which is at tilt angle is φ = 64◦ ,

remains invariant under change of density. In the second example, we have one more

small particle, i.e. for C = 2/3, at the same tilt angle, the system changes slightly over

different values of density, but maintain essentially the same pattern, Fig. 6.9(b), (f )

and (j). The reason is that at high density, the excluded volume becomes significant

and reduces the space between the particles. Therefore, the small particles are unable

to maintain their pattern they had at lower densities in order to take advantage of

the anisotropy.

6.4 Conclusions 106

In the first two examples, anisotropy does not affect the system’s scale-free fea-

tures. However, this is true so long there is a repulsive interaction among the particles

along the x-axis. At φ = 54.7◦ there is no interaction anymore between the particles

in the x-direction, which breaks the scale-free feature. As seen at Fig. 6.9(c), (g) and

(k), the pattern may essentially remain the same, but along the x-axis the particles

ceases to scale by not changing positions as density changes.

A final example is displayed at Fig. 6.9(d), (h) and (l), where the composition of

the system C = 1/3, which has higher packing fraction than C = 2/3, compare with

Fig. 6.9(b), (f) and (j). The anisotropy, similar to the case C = 2/3, does not affect the

system being scale-fee, till the point there is no interaction in one direction, as it

happens for φ = 54.7◦ .

In the final block of results, we look at structure with composition C > 1/2, but

with smaller diameter ratio, z = 0.3296. From a quick look at the structures, the

same behavior is observed when z = 0.5; initially the structures are perturbed, then

new structures emerge, and at φ = 54.7◦ chains are formed. Despite the similarities

there are some small but significant differences. The first one occurs for C = 2/3

at φ = 54.7◦ , Fig. 6.10(e) where the small particles form a chain, but they are not

equally spaced; the small particles form pairs. The reason behind this is that the

small particles attempt to maximize their distance from the big particles not only

along the y-axis but also the x-axis due to the absence of interactions between the

particles along the x-axis. This effect is not observed for z = 0.5, because two small

particles to cover the distance between two particles of the primary lattice.

A smaller diameter ratio also has an effect of reducing the number of chains

formed by the small particles at φ = 54.7◦ . For example, comparing C = 3/4 at z = 0.5

(Fig. 6.6(o)) and at z = 0.3296 (Fig. 6.10(j)), we observe that at lower diameter ratio

there is one less chain. As mentioned above this happens because more particles can

fit into one chain without steric repulsive iteractions come into effect.

6.4 Conclusions

We have applied genetic algorithms to study equilibrium configurations of binary

mixture of two dimensional dipolar particles in the presence of a tilted external mag-

netic field, which introduces anisotropic interactions between the particles. At small

degree of anisotropy the structures are perturbed, while once the anisotropy reaches

the maximum degree, that is when there is no repulsive (or attractive) interactions

90◦ 74◦ 64◦ 59◦ 54.7◦

C = 2/3

6.4 Conclusions

C = 3/4

C = 4/5

C = 5/6

Figure 6.10: (Color online) Minimum energy configurations for C ≥ 1/2, ρ = 0.6, and z = 0.3296.

107

6.4 Conclusions 108

along the direction of the magnetic field, the particles form chains aligned parallel to

the field. By varying the degree of anisotropy and density we found a rich spectrum

of new structures.

Chapter 7

Conclusions

This thesis is divided into three parts. The first part contains two studies of ph-

ysisorbed polymers (linear and star) on surfaces. In the second part, an attempt has

been made to realise a scale-free near-neighbour network in computer simulations

of polydisperse fluids of hard disks. In the final part we explore the minimum en-

ergy configurations of two dimensional binary mixture of magnetic colloids in tilted

external fields.

In the first part, we were motivated by atomic-force microscopy images of poly-

mer films (either linear chains or stars), on mica surfaces, adsorbed from dilute

solution following a good solvent-to-bad solvent quenching transition. In the case

of linear chains a bimodal cluster distribution is observed under certain experimen-

tal conditions. We used a bead-spring molecular model and Langevin dynamics

computer simulations and demonstrated that a bimodal distribution can be repro-

duced in the simulations and rationalised on the basis of the polymer structures

prior to the quench. In addition to providing insight on experimental observations,

the simulation results support a number of predicted scaling laws such as the decay

of the monomer density as a function of distance from the surface. Additionally,

the same methodology has been used to study many arms stars polymers, in which

the soft-colloidal nature of star polymers was studied in the proximity of a surface.

Measurements of the fraction of monomers bound to the surface, height, radius

of gyration and asphericity provided valuable insights on the structures. We ratio-

nalised the results and allowed us to map the stars’ behaviours (from polymer-like to

soft-colloid) based on the number of arms and their length on a given surface.

In the second part of this thesis we examine the neighbour network in a two-

dimensional polydisperse hard-disk fluid with diameter distribution p(σ) ∼ σ−4 is

examined using constant-pressure Monte Carlo simulations.Graphs are constructed

from vertices (disks) with edges (links) connecting each vertex to k neighbouring

109

7 Conclusions 110

0.36, the decay of the network degree distribution is observed to be consistent with

the power-law k −γ where the exponent lies in the range 5.6 ≤ γ ≤ 6.0. Comparisons

with the predictions of a maximum-entropy theory suggest that this apparent power-

law behaviour is not the asymptotic one, and that p k ∼ k −4 in the limit k → ∞. This

asymptotic dependence is consistent with the simple idea that for large disks, the

number of neighbours is proportional to the disk diameter. A power-law decay of

the network degree distribution is one of the characteristics of a scale-free network.

The assortativity of the network is measured, and is found to be positive, meaning

that vertices of equal degree are connected more often than in a random network.

Finally, the equation of state is determined and compared with the prediction from

a scaled-particle theory. Very good agreement between simulation and theory is

demonstrated.

In the final part, we use genetic algorithms, which allow to efficiently search

energy landscapes and obtain their global minimum, to investigate the ordered

equilibrium structures formed by binary mixtures of anisotropic dipolar particles

confined to a plane, under the presence of an external magnetic field. Anisotropy

is introduced by tilting the external magnetic field with respect to the interface.

Initially, as the field is tilted the isotropic structures are perturbed only slightly, but

as anisotropy becomes enhanced by tilting the magnetic field further new structures

emerge. Once anisotropy becomes dominant, the structures form chains parallel to

the field.

Appendix A

experiment, theory and computer simulation, A. O. Ivanov, S. S. Kantorovich,

E. N. Reznikov, C. Holm, A. F. Pshenichnikov, A. V. Lebedev, A. Chremos, and P. J.

Camp, Phys. Rev. E 75, 061405 (2007).

on surfaces, E. Glynos, A. Chremos, G. Petekidis, P. J. Camp, and V. Koutsos,

Macromolecules 40, 6947-6958 (2007).

assortativity, A. Chremos and P. J. Camp, Phys. Rev. E 76, 056108 (2007).

experiment, theory, and computer simulations, A. O. Ivanov, S. S. Kantorovich,

E. N. Reznikov, C. Holm, A. F. Pshenichnikov, A. V. Lebedev, A. Chremos, and P. J.

Camp, Magnetohydrodynamics 43, 393-400 (2007).

ulation study, A. Chremos, E. Glynos, V. Koutsos, and P. J. Camp, Soft Matter,

accepted (2008).

mopolymer on surfaces: an atomic force microscopy study, E. Glynos, A. Chre-

mos, P. J. Camp, E. Theofanidou, and V. Koutsos, submitted to Macromolecules.

tion study, A. Chremos, E. Glynos, V. Koutsos and P. J. Camp, in preparation.

tilted external fields, A. Chremos and C. N. Likos, in preparation.

111

A Appendix - Papers published 112

A. Chremos, and P. J. Camp, in preparation.

Bibliography

[4] S. M. Ilett, A. Orrock, W. C. K. Poon, P. N. Pusey, Phys. Rev E 51, 1344 (1995).

[7] M. E. J. Newman, Proc. Natl. Acad. Sci. U.S.A. 98, 404 (2001).

[8] R. Aveyard, B. P. Binks, J. H. Clint, et al., Phys. Rev. Lett. 88, 246102 (2002).

Oxford, 1987).

[10] J.-P. Hansen, I. R. McDonald, Theory of simply liquids (Academic Press, London,

2006).

London, 2001).

Cornell, 1979).

2003).

London, 1983).

(1999).

113

BIBLIOGRAPHY 114

[19] K. B. Migler, H. Hervet, L. Leger, Phys. Rev. Lett. 70, 1993 (1993).

[22] E. Pludeman, N. Collins, Adhesion Science and Technology (Plenum Press, New

York, 1975).

33, 2573 (2000).

Macromolecules 30, 4719 (1997).

32, 1233 (1999).

[33] K. Binder, A. Milchev, J. Baschnagel, Annual Reviews Mater. Sci. 107, 107 (1996).

[35] D. Mukherji, G. Bartels, M. H. Müser, Phys. Rev. Lett. 100, 068301 (2008).

(2008).

[40] R. Descas, J.-U. Sommer, A. Blumen, J. Chem. Phys. 125, 214702 (2006).

[41] S. Metzger, M. Müller, K. Binder, J. Baschnagel, J. Chem. Phys. 118, 8489 (2003).

(1996).

BIBLIOGRAPHY 115

[47] O. Guiselin, L. T. Lee, B. Farnoux, A. Lapp, J. Chem. Phys. 95, 4632 (1991).

Lett. 56, 200 (2001).

(Oxford University Press, Oxford, 2004).

R481 (2004).

[60] E. Eisenriegler, Polymers Near Surfaces (World Scientific, River Edge, 1993).

BIBLIOGRAPHY 116

[76] K. Shida, K. Ohno, M. Kimura, K. Y., J. Phys. Chem 105, 8929 (1996).

(2008).

Berlin, 1984).

Matt. 13, 6177 (2001).

[89] M. Faloutsos, P. Faloutsos, C. Faloutsos, Comptut. Commun. Rev. 29, 251 (1999).

[90] H. Jeong, B. Tombor, R. Albert, Z. N. Oltvai, A.-L. Barabási, Nature 407, 651

(2000).

(2001).

BIBLIOGRAPHY 117

Phys. Rev. E 64, 041902 (2001).

[104] J. S. Andrade, Jr., J. Herrmann, R. F. S. Andrade, L. R. da Silva, Phys. Rev. Lett. 94,

018702 (2005).

[118] M. A. Pshkin, K. J. Strandburg, N. Rivier, Phys. Rev. Lett. 67, 1803 (1991).

BIBLIOGRAPHY 118

[127] M. P. Allen, G. T. Evans, D. Frenkel, B. M. Mulder, Adv. Chem. Phys. 86, 1 (1993).

[128] J.-P. Hansen, R. M. I, Theory of simple liquids (Academic Press, London, 1986).

Europhys. Lett. 20, 559 (1992).

[138] K. Zahn, A. Wille, G. Maret, S. Sengupta, P. Nielaba, Phys. Rev. Lett. 90, 155506

(2003).

[141] N. Hoffmann, F. Ebert, C. N. Likos, H. Löwen, G. Maret, Phys. Rev. Lett. 97,

078301 (2006).

[146] V. A. Froltsov, R. Blaak, C. N. Likos, H. Löwen, Phys. Rev. E 68, 061406 (2003).

(Oxford University Press, Clarendon, 1981).

BIBLIOGRAPHY 119

[148] J. H. Holland, Adaption in Natural and Artificial Systems (MIT Press, 1992).

(Springer-Verlag, 1992).

Learning (Addison-Wesley, Reading, 1989).

- HSpeed NotesUploaded bymohdrashdansaad
- 2 9Uploaded byanon-29683
- 1bcape2001unit2paper2.pdfUploaded bykesna
- Solution Manual for Fundamentals of Modern VLSI Devices 2n Ed - Yuan Taur, Tak NingUploaded byMasoomeh Akbarzadeh
- slide_12.pdfUploaded byannisya
- Entrance Loss CoefficientsUploaded bytitibi
- Flow NetsUploaded byMAWAR08
- ijcerv5n4spl_01Uploaded byfaisalwalksicloud
- Channel 3Uploaded byparandhamaiah
- Quiz 4Uploaded byEngr Russell Sabir
- BARRA HUECA IMPRIMIR.docxUploaded byDaniela
- dipseUploaded byPatrick Sibanda
- 18-03-Anskey.pdfUploaded byNagendra Bharadwaz
- Rizaldi_pipeflowUploaded byRizaldi Riz
- Hogy Számol a Külföldi ExtrudertUploaded byHardi I
- 68 Lot Phase 5 Answer KeyUploaded byVivek Chauhan
- Calculating the concentration of a substanceUploaded byJuan Pa Valverde
- 0625_w01_ms_2 (1)Uploaded byahmed
- Examination 1 Soln (9 23 2013)Uploaded byKevin
- Observe.chem.Changes8Uploaded byctremblaylcsd150
- RQ1Articulo for Mole Problems Call AvogadroUploaded byKaren Rodriguez
- Examination 1 SolnUploaded byAnonymous BYXK2W
- Vitalii N. Chukov. Surface Roughness Structure and Rayleigh Scattering Laws of Rayleigh WaveUploaded byVitalii-N-Chukov
- rr1Uploaded byapi-3701628
- Vladimir V. Riabov- Heat Transfer on a Hypersonic Sphere with Diffuse Rarefied-Gas InjectionUploaded byFraosm
- Handbook_of_Magnetic.pdfUploaded byQassem Mohaidat
- Development of Novel PH-sensitive Thiolated ChitosanPMLA NanoparticlesUploaded byArmando Lerma
- DATA exp.docxUploaded byRhyea Engay
- 2physical Properties of RockUploaded bySheshu Babu
- chemistry lab reportUploaded byapi-334420312

- Ineffective Assistance of Counsel and Failure to Investigate 08-22-07Uploaded byHouston Criminal Lawyer John T. Floyd
- Oct14_catUploaded bySaraju Nandi
- essaypeerreview 1Uploaded byapi-349319202
- a study on customer awareness bajaj allianzUploaded byManjunath@116
- Shaw & Adam_alberto Fortis and the Istrian Karst, Croatia, In 1770 and 1771Uploaded byCaegeo
- Company Profile - CTEUploaded bymishraavneesh
- Civil SyllabusUploaded byShah Khan Mps
- Engine BreathingUploaded byNimalan
- Is 6623 High Strength Structural NutsUploaded byprashantlingayat
- App Fabric Web FarmUploaded bymumerdraz
- Listening Tip IELTS Test AdviceUploaded bymysteriousm333
- EAPPUploaded byButterfly 0719
- A Regulatory Framework for MicrogridsUploaded byYuliana Liu
- Gear MeasurementUploaded byAnand Babu
- Servqual Model Spar Hyper 01Uploaded byPrudhvinadh Kopparapu
- Bq Magazine - October 2013Uploaded bybq magazine
- 2011 Sitcom FurnitureUploaded bycalligariscollection
- AI_2008_3_1_Tellez_etal.pdfUploaded byRoxy Destiny
- Technotex FairUploaded byarun_c86
- 7-daniel-cáceres---live-tank-circuit-breakers.pdfUploaded byElmer UNI
- AMNH Annual Report 2017Uploaded byc brouillette
- pengaturcaraan komputerUploaded byKina Ibrahim
- 1 2 7 p understandingdigitaldesign rng(finished)Uploaded byapi-287657167
- Dr Andres Ayuela FernandezUploaded byjovmicic
- Cubic EquationUploaded bymacynthia26
- Stop The IndecisionUploaded byFarhan Hayat
- Career Experts and Bloggers Unite to Help Job-Seekers in Third Annual Job Action DayUploaded byPR.com 2
- Accumulation of Capital in Southern AfricaUploaded byabhayshk
- Virata ECE221 SolutiontoDifferenceEquationUploaded byJeremy Lorenzo Teodoro Virata
- student sample b jamestown 1Uploaded byapi-414159283